Hindsight check: LL1-17 (Ch 17 Conclusions)#

Source section: EEA, Late lessons from early warnings: the precautionary principle 1896–2000, Environmental Issue Report No 22 (2001), Chapter 17 “Conclusions”, report pp. 192–194. Check window: publication (2001) to late September 2026. Checked: 25 September 2026.

Method note. - General web search was unavailable for this pass (the session’s search budget was exhausted). I retrieved sources by fetching known primary repositories directly: EUR-Lex (legislation and CJEU/General Court judgments), the Stockholm Convention Secretariat (pops.int), the UNEP Ozone Secretariat, the European Parliament’s Legislative Observatory and research service (EPRS), UK Hansard (via the Parliament API), GOV.UK (OPRED), OSPAR, WHO, FDA, the National Academies, and the Europe PMC, Crossref and OpenAlex bibliographic databases for peer-reviewed abstracts. - Some pages refused automated retrieval: ECHA (web firewall), Nature, The Lancet, SSRN and the University of Chicago journals site (403), CURIA (redirect loop), and hansard.parliament.uk itself (403; the Hansard API worked). Where a point rests on a source I could not open, or on a secondary source, I say so. - The 2013 volume. The EEA publication page for Late lessons from early warnings: science, precaution, innovation (EEA Report 1/2013) now serves only the 48-page summary; the DTU repository copy of the full report returned 403. So I could not re-check the 2013 annex that summarises the 2001 cases. Ch 17 is a conclusions chapter, not a case, so the relevant 2013 update is how the second volume treats the twelve lessons and the chapter’s framework. That is covered below from the 2013 summary. - I read only the digest and source extract for this section, plus web sources.

Overview#

Ch 17 makes two kinds of claim. Twenty-five years on they have fared very differently.

The diagnostic claims have mostly held or strengthened. - Warnings ignored, appraisal too narrow, alternatives not weighed (p. 192). This is borne out by the 2013 volume’s nine new “false negative” cases and by later cases the EEA did not select. - Industry documents show that producers of per- and polyfluoroalkyl substances (PFAS) knew of toxicity by 1970, “forty years before the public health community” (Gaber et al. 2023). - In 2023 EFSA cut the tolerable daily intake for bisphenol A (BPA) to 0.2 ng/kg bw/day, 20,000 times lower than its 2015 temporary value. The critical effect had moved to the immune system, an endpoint absent from earlier appraisals. The EU then banned BPA in food-contact materials (Reg. (EU) 2024/3190). - Substitutes adopted without appraising alternatives created new problems: hydrofluorocarbons after CFCs, needing the 2016 Kigali Amendment; organophosphate flame retardants after polybrominated diphenyl ethers (PBDEs). - The main caveat still applies: cases are selected on outcome. Some warnings the EEA promoted in 2013 have not been borne out. A WHO-commissioned systematic review found no association between mobile-phone use and brain tumours (Karipidis et al. 2024). - Level of proof is a political choice that distributes the costs of error (p. 193). The EU General Court confirmed this in Pfizer v Council (11 September 2002): the scientific committee had “neither democratic legitimacy nor political responsibilities”, so the Council could disregard its conclusions (paras 200–201). Several regimes now tie graded strengths of evidence to consequences. EU pesticide approval sets hazard “cut-offs” at carcinogen, mutagen and reprotoxin categories 1A and 1B. The US Toxic Substances Control Act, as amended in 2016, requires risk evaluation “without consideration of costs or other nonrisk factors”. - Persistence and bioaccumulation as predictors of harm, and long-term monitoring (Table 17.1). These strengthened most of all as policy. - The Stockholm Convention entered into force in 2004. It has grown from 12 to 37 listed entries, including four PFAS families. Its 2023 effectiveness evaluation found concentrations of the initial POPs declining in air and people. - The EU wrote persistence, bioaccumulation and mobility into hazard classification in 2023 (Delegated Reg. (EU) 2023/707). - Monitoring caught two post-2001 surprises: PFOS in wildlife worldwide (Giesy & Kannan 2001) and illegal CFC-11 emissions (Montzka et al. 2018). - Antibiotic growth promoters as “precautionary prevention” (Table 17.1). This is the clearest case of vindication. - The EU ban took effect on 1 January 2006 as scheduled. The court upheld the 1998 precautionary withdrawal of virginiamycin in 2002. - Growth-promotion use was later banned for veterinary medicines generally (Reg. (EU) 2019/6, Art. 107(2)) and applied to imports from 3 September 2026. - A 2017 meta-analysis links restrictions to less resistance in animals and in people who work with animals. - When China banned colistin as a growth promoter in 2017, colistin-resistant E. coli in pig faeces fell from 34.0% to 5.1%. Human carriage of the mcr-1 resistance gene fell from 14.3% to 6.3%.

The prescriptive and optimistic claims remain contested or untested. - Precaution “stimulat[es] more innovation” (p. 194). This is not established. - The best evidence on regulation and innovation supports the weak Porter hypothesis: regulation induces some clean innovation. The strong version, that regulation improves competitiveness, is not supported. - The largest meta-analysis finds “the most likely scenario is statistical insignificance” (Cohen & Tubb 2018). - The EU now carries an “innovation principle” in the recitals of its research framework regulation (Reg. (EU) 2021/695, recital 6). Industry groups promoted it explicitly as a counterweight to precaution. - Stakeholder involvement adds information and trust without stifling innovation (p. 193). Only partly supported. - The UK’s 2003 GM Nation? debate was flawed in design and self-selected. - Experimental evidence shows mini-publics raise perceived legitimacy mainly when their recommendations are honoured (Germann et al. 2022). - No study I found tests the innovation half of the claim. - Controversies turn on values (p. 193). Supported in part. - EU law now lets member states restrict genetically modified (GM) crop cultivation on socio-economic, land-use and “public policy” grounds (Directive (EU) 2015/412). - The US National Academies found “no substantiated evidence” that GM foods are less safe, and attributed national regulatory differences to “social, political, legal, and cultural differences”. - The Brent Spar case is messier than the chapter implies. The outcome rested partly on an oil estimate by Greenpeace that was wrong and was withdrawn. The UK’s 1996 scientific review found the impacts of deep-sea disposal “very small”. The value judgement against dumping was then codified in OSPAR Decision 98/3. - False positives are rare (p. 194, implicitly). Only four “genuine” regulatory false positives were found among 88 alleged cases (Hansen et al. 2007). But the criteria were contested (Cox 2007), and the method does not count alarms that were never regulated. Risk-for-risk trade-offs are real in the section’s own flagship case: therapeutic antibiotic use rose after the growth-promoter ban, although total use stayed far below its earlier peak. - Institutions embracing the twelve lessons would have fewer or cheaper surprises (p. 194). This is untested. No institution adopted the lessons as a package, and no evaluation compares adopters with non-adopters. - A random-sample comparison found no significant overall difference in precaution between the EU and US over 1970–2004 (Hammitt et al. 2005). - The jurisdiction with precaution in its treaty still faced large PFAS and BPA exposure “surprises”. EFSA found parts of the European population above its PFAS tolerable weekly intake (2020), and dietary BPA exposure two to three orders of magnitude above the new tolerable daily intake (2023).

Table 17.1’s typology was partly taken up. It fed into European Parliament research-service guidance and into the risk-governance literature. But its most influential descendant, Stirling’s four-way scheme of risk, uncertainty, ambiguity and ignorance (2007, 2010), comes from one of the 2001 editors and adds a category the 2001 table lacked. The coinage “precautionary prevention” appears in the European Parliament research service’s glossary definitions but not, in the sources I checked, in legislation.

Implication for weight. Treat Ch 17’s diagnosis as strong, with a clear evidence base that post-2001 events have added to. The strongest parts are the claim that evidence thresholds are political choices about who bears the cost of error, and the case for property-based screening and monitoring when the specific harm is unknown. Treat its optimism about innovation, trust and the twelve lessons as a package as advocacy that remains unproven. The 2013 volume restated the twelve lessons unchanged and asserted “growing evidence” on innovation. That comes from the same editorial stable, so it is not independent confirmation.

Claim-by-claim#

Claim 1: Warnings ignored, appraisal too narrow, actions taken without weighing alternatives (p. 192)#

Original claim (p. 192). The case studies “provide many examples where ‘early warnings’, and even ‘loud and late’ warnings, were clearly ignored; where the scope of hazard appraisal was too narrow; and where regulatory actions were taken without sufficient consideration of alternatives, or of the conditions necessary for their successful implementation in the real world.”

Subsequent developments

The EEA’s own follow-up (2013). - The second volume examined nine new false-negative cases: lead in petrol, perchloroethylene (PCE) in water, Minamata, beryllium, environmental tobacco smoke, vinyl chloride, dibromochloropropane (DBCP), BPA and DDT. It reports “three common themes”: “more than sufficient evidence for much earlier action; slow and sometimes obstructive behaviour by businesses …; and the value of independent scientific research and risk assessments” (EEA 2013 summary, p. 10). - It judged that “by and large, societies are not making the most use of the costly lessons that can be gleaned from their histories” (p. 10). - It restated that “the 14 case studies and 12 key lessons from the 2001 report remain highly pertinent today” (p. 9). - These cases were again chosen by the EEA’s editors and advisory board (p. 9), so they share the 2001 selection-on-outcome limit.

Later cases outside the EEA selection. - PFAS (warnings ignored). Gaber, Bero & Woodruff (2023), reviewing previously secret industry documents, found that companies knew PFAS were “highly toxic when inhaled and moderately toxic when ingested” by 1970, “forty years before the public health community”. They document strategies to suppress unfavourable research and distort public discourse. Separately, persistence-driven global contamination was first reported through wildlife monitoring in 2001 (Giesy & Kannan 2001). By 2020, EFSA concluded that parts of the European population exceeded its new tolerable weekly intake for four PFAS, with effects on the immune system as the critical endpoint. - BPA (appraisal too narrow; also a 2013 case). - EFSA’s 2023 re-evaluation set a tolerable daily intake of 0.2 ng/kg bw/day, against a 2015 temporary value of 4 μg/kg bw/day. The critical effect was on immune (Th17) cells, not the reproductive and endocrine endpoints that dominated earlier debate. EFSA found that dietary exposure in all age groups exceeded the new intake “by two to three orders of magnitude”. - The Commission then prohibited BPA in food-contact materials (Reg. (EU) 2024/3190, 19 December 2024), with 18- and 36-month transitions. - In both PFAS and BPA, the endpoint that finally drove action lay outside the scope of earlier appraisals. That is direct later support for the “too narrow” diagnosis. - Neonicotinoids (a 2013 case; action eventually taken). - EFSA identified “high acute risks for bees” from imidacloprid for most crops (conclusion of 13 October 2016). The EU then confined imidacloprid to permanent greenhouses (Implementing Reg. (EU) 2018/783, 29 May 2018), roughly two decades after French beekeepers’ first reports in 1994 (as summarised in EPRS 2015). - In January 2023 the Court of Justice held that member states could not use emergency authorisations to override that ban (Case C-162/21, 19 January 2023). - Actions without weighing alternatives (regrettable substitution). - CFCs to HFCs. The Montreal Protocol’s success encouraged hydrofluorocarbons, potent greenhouse gases that had to be phased down by the Kigali Amendment (agreed October 2016, in force 1 January 2019). - PBDEs to organophosphate flame retardants. After PBDEs were listed under the Stockholm Convention, organophosphate esters were adopted “despite a lack of comprehensive understanding of their ecological and health consequences” and are now widespread (Qadeer et al. 2024). - Learning from this. The EU’s 2024 BPA regulation explicitly anticipates substitution. It bars other bisphenols with harmonised hazard classifications, including bisphenol S (classified as toxic to reproduction, category 1B), without an up-to-date EFSA assessment (Reg. (EU) 2024/3190, recitals).

Evidence that cuts the other way. - Mobile phones (2013 volume, Ch 21). - The 2013 volume argued for “precautionary actions now to reduce head exposures” on the basis of case-control studies and IARC’s 2011 Group 2B classification (EEA 2013 summary, p. 31). - A WHO-commissioned systematic review later pooled 63 studies (1994–2022). It found mobile-phone use “was not associated with an increased risk of glioma (mRR = 1.01, 95% CI 0.89–1.13), meningioma … acoustic neuroma”, with no dose-response with cumulative call time (Karipidis et al. 2024). - An early warning the EEA foregrounded has, on current evidence, not been borne out. - GM crops (2013 volume, Ch 19). The 2013 chapter framed GM crops as ill-suited to sustainable food security. The US National Academies (2016) found “no substantiated evidence that foods from GE crops were less safe than foods from non-GE crops”, and no adverse effects in long-term livestock health data. - These do not falsify Ch 17’s claim, which concerns warnings that proved correct. They show that a list of vindicated warnings is not a base rate. Not every early warning is right, and the reports do not show how to tell which will be.

Verdict: held up. It is strengthened for persistent chemicals and for substitution hazards, and it still carries the selection caveat.

Implications for weight. The three failure mechanisms recur in cases outside the EEA’s selection, sometimes in stronger form: - ignoring or suppressing warnings - appraisal scoped too narrowly (the endpoint that matters lies outside the frame) - intervention without appraising alternatives (regrettable substitution)

They can be used with confidence as a lens. What cannot be drawn from Ch 17 is a claim that warnings are usually right. The 2013 volume’s own record includes warnings that did not hold up.

Claim 2: The risk / uncertainty / ignorance typology will clarify the precaution debate (p. 192, Table 17.1)#

Original claim (p. 192). “What is sometimes loosely referred to as ‘uncertainty’ mixes up the analytically distinct concepts of ‘risk’, ‘uncertainty’ and ‘ignorance’.” Pairing these with “prevention”, “precautionary prevention” and “precaution” (Table 17.1) is offered as “a contribution to clarifying the meaning of six basic concepts”.

Subsequent developments

Scholarly uptake, largely through one of the 2001 editors. - Stirling (2007, EMBO Reports) developed a four-way scheme: - risk: outcomes and probabilities both well characterised - uncertainty: outcomes known, probabilities not - ambiguity: the outcomes themselves are contested (“ecological, agronomic, safety, economic or social criteria” in the GM case) - ignorance: neither outcomes nor probabilities can be characterised - He presented precaution as “not a complete decision rule” but a normative guide that triggers broader appraisal. Nature’s “Keep it complex” (Stirling 2010, 468:1029–1031, 23 December 2010) carried this to a wide audience. It argues that under uncertainty, ambiguity and ignorance, expert advice should be “plural and conditional” (as summarised by EPRS 2015, p. 21; the Nature page itself was paywalled). - Stirling was on the 2001 editorial team (see digest). So the most influential development of Table 17.1 is a revision by one of its authors. The addition of ambiguity concedes a gap in the 2001 table: disagreement about which outcomes matter and how to frame them is different from not knowing probabilities. - Ironically, Ch 17’s own text makes that point (controversies “have much to do with public values”, p. 193) but does not build it into the table. - Risk-analysis scholars took up the problem from another direction. Aven (2011, Risk Analysis 31:1515–1525) argued that “scientific uncertainty” in precautionary definitions is unclear. He held that “it is not sufficient that the threats or hazards are uncertain. A stronger requirement is needed”, and proposed a new classification. - The International Risk Governance Council’s framework (IRGC 2005, Renn) classifies risk problems by complexity, uncertainty and ambiguity, and matches stakeholder involvement to each. I confirmed the framework and its 2005 origin on the IRGC site, but not the category wording, which the page did not show.

Policy uptake: advisory, not legal. - The European Parliament’s research service built its 2015 briefing on the precautionary principle around these distinctions. Its glossary defines ignorance as impacts unknown “with the result that the probabilities are also unknown; may lead to precautionary measures to minimise the impact of surprises”. It defines uncertainty as impacts “likely but the probabilities are unknown; may lead to precautionary measures to reduce exposure”. Both track Table 17.1 closely. The briefing’s Figure 1 reproduces Stirling’s 2007 four-way matrix (EPRS 2015, pp. 3, 7). - The EEA’s own 2013 working definition invokes “situations of scientific complexity, uncertainty and ignorance” and “an appropriate strength of scientific evidence” (EEA 2013, p. 649, as quoted in EPRS 2015, p. 10). It swaps in complexity rather than adopting ambiguity. - I found no EU legislative act or Commission guidance that adopts Table 17.1’s six-term scheme. The Commission’s 2000 Communication remains the reference for EU precaution (EPRS 2015, pp. 8–9) and has not been revised. The General Court’s 2002 Pfizer judgment works with the Communication’s risk-assessment and risk-management vocabulary, not the EEA’s (T-13/99, para. 119). - I did not find the term “precautionary prevention” in any legal instrument. This rests on the documents checked here, not an exhaustive search.

Did it clarify the debate? The EPRS still reported in 2015 that “there is no universally accepted definition of the precautionary principle”, and that the main axis of disagreement is “the degree of scientific uncertainty likely to lead to action” (pp. 6–7). The typology gave analysts a shared vocabulary. It did not settle the definitional dispute it was meant to address.

A dynamic the table implies but later evidence makes visible. Knowledge states move, and they move unevenly across a single issue. Table 17.1 dates antibiotics in feed as “uncertainty” from 1969 “to the present”. By 2024 the joint surveillance report of the European Centre for Disease Prevention and Control (ECDC), EFSA and the European Medicines Agency (EMA), known as JIACRA, had quantified associations between animal consumption and resistance “for certain combinations of bacteria and antimicrobials” (JIACRA IV, 2024). Parts of that issue have moved towards “risk” while others remain uncertain.

Verdict: partly held up. The distinction became standard in precaution scholarship and in European Parliament advisory work. It was not adopted in law. Its most influential successor added ambiguity, a category the 2001 table lacked.

Implications for weight. The core move of Table 17.1 is sound and widely accepted: separating not-knowing-probabilities from not-knowing-outcomes, and matching responses to each. As a lens, though, it should be used in its later four-way form. The 2001 three-way version folds disputes about framing and values into “uncertainty”. Any use should also treat knowledge states as specific to each sub-question and changing over time, not fixed for a whole technology.

Claim 3: Persistence and bioaccumulation as predictors, broad long-term monitoring, and technological diversity are effective responses to ignorance (p. 192, Table 17.1)#

Original claim (p. 192, Table 17.1). Precaution in the face of ignorance means “use of properties of chemicals such as persistence or bioaccumulation as ‘predictors’ of potential harm; use of the broadest possible sources of information, including long term monitoring; promotion of robust, diverse and adaptable technologies and social arrangements … with fewer technological ‘monopolies’ such as asbestos and CFCs.”

Subsequent developments

Property-based screening: widely institutionalised. - Stockholm Convention. - Adopted 22 May 2001 and in force since 17 May 2004. It lists chemicals on the basis of persistence, bioaccumulation, potential for long-range transport and adverse effects (pops.int, overview). - It began with 12 “initial POPs”. The Secretariat’s list now shows 33 entries in Annex A (elimination), 2 in Annex B (restriction: DDT and PFOS) and 2 more that appear only in Annex C (unintentional production: dioxins and furans). That makes 37 listed entries, many covering groups of substances (pops.int, “All POPs”, retrieved 25 September 2026). - Additions came at every Conference of the Parties from 2009 onward. Among them are four PFAS families: PFOS (2009), PFOA (2019), PFHxS (2022) and long-chain perfluorocarboxylic acids (COP-12, 28 April to 9 May 2025, alongside chlorpyrifos and medium-chain chlorinated paraffins) (pops.int, “The New POPs”). - The second effectiveness evaluation (welcomed at COP-11, 2023) found: “Monitoring results indicate that regulations targeting POPs have succeeded in reducing levels of POPs in humans and the environment. For the initial POPs, concentrations measured in air and in human populations have declined … For the newly listed POPs, concentrations are beginning to show decreases, although in a few instances, increasing and/or stable levels are observed.” - The same evaluation notes that all the Convention’s mechanisms are in place “with the exception of procedures and mechanisms on compliance”, and that limited national data remains “a major challenge” (pops.int, Effectiveness Evaluation, 2023 outcomes). - EU law. - The pesticides regulation makes persistence-based criteria absolute bars. An active substance “shall only be approved where it is not considered to be a persistent organic pollutant” or “a persistent, bioaccumulative and toxic (PBT) substance” (Reg. (EC) 1107/2009, Annex II, points 3.7.1–3.7.2). - In 2023 the Commission added hazard classes for endocrine disruption, PBT/vPvB (very persistent, very bioaccumulative) and PMT/vPvM (persistent, mobile and toxic; very persistent, very mobile) to the classification regulation. It based these on “experience and increased scientific knowledge gained in identifying substances as … PBT … vPvB … PMT … and vPvM” under the EU chemicals regulation, REACH (Delegated Reg. (EU) 2023/707, recitals 3, 7). - Adding mobility extends the 2001 logic, which named only persistence and bioaccumulation. - Towards persistence alone. Cousins et al. (2019) argue that “high persistence alone should be established as a sufficient basis for regulation”. Continuous release of a highly persistent chemical leads to “continuously increasing contamination irrespective of the chemical’s physical-chemical properties”, and reversal takes “decades, centuries or even longer”. Cousins et al. (2020) applied this to PFAS “as a chemical class”. - This is the most direct later elaboration of Table 17.1’s “predictors of potential harm” idea: act on the property, before the specific harm is known. - In 2023 five European states proposed restricting PFAS as a class under REACH on largely this rationale. The ECHA pages documenting the proposal and its status refused automated retrieval, so I have not verified its details or current stage here.

Limits exposed after 2001. - Speed. Listing under Stockholm takes years per chemical and often comes “with specific exemptions” (for example PFOS in 2009 “with eight acceptable purposes and 12 specific exemptions”; pops.int, 2023 highlights). - Substitution. Screening one substance or family at a time can push use towards unscreened analogues. PBDEs were followed by organophosphate esters (Qadeer et al. 2024), and PFOS and PFOA by other PFAS. This is part of what drives the class-based argument. - Compliance. No compliance mechanism existed as of the 2023 evaluation.

Long-term monitoring: vindicated twice over. - PFOS was discovered to be globally distributed through wildlife sampling “from the Arctic and the North Pacific Oceans” (Giesy & Kannan 2001, Environ Sci Technol 35:1339–1342). - Atmospheric monitoring detected “an unexpected and persistent increase” in CFC-11 emissions after 2012 of about 13 ± 5 Gg per year, “despite reported production being close to zero since 2006”, which pointed to “unreported new production” (Montzka et al. 2018, Nature 557:413–417). This is precisely the surprise-detecting function Table 17.1 assigns to monitoring. - The China colistin study (see Claim 4) likewise closes by calling for continued monitoring “to act as an early warning system”.

Technological diversity and fewer “monopolies”: little uptake, no test. - Stirling developed a general framework for measuring diversity in science and technology portfolios (Stirling 2007, J R Soc Interface 4:707–719). - The 2013 volume argued for “diversity of locally sensitive technological options” in food and energy (EEA 2013 summary, p. 11). - I found no regulatory regime that uses technological diversity as an explicit precautionary criterion, and no empirical test of whether diversity reduced the cost of surprises. This element remains a plausible hypothesis.

Verdict: strengthened for property-based screening and for monitoring, which have been widely institutionalised and have reduced exposures or detected surprises. Unclear for technological diversity.

Implications for weight. This is the section’s best-supported practical lesson. It is technology-neutral: when the specific harm is unknown, screen on properties that predict the scale and irreversibility of harm if something goes wrong, and fund the monitoring that will reveal what screening misses. Persistence, bioaccumulation, mobility and reach are such properties. Three qualifications from later experience should travel with the lesson: - screening one substance at a time invites substitution - listing is slow and exemption-laden - without compliance mechanisms, detection does not guarantee correction

The diversity recommendation should be cited as an untested proposal.

Claim 4: Reducing or eliminating human exposure to antibiotics in animal feed is appropriate “precautionary prevention” (p. 192, Table 17.1)#

Original claim (p. 192, Table 17.1). Under “uncertainty” (“known impacts; unknown probabilities”, 1969–present), the appropriate action is “precautionary prevention”, for example “reduce/eliminate human exposure to antibiotics in animal feed”.

Subsequent developments

The EU phase-out was completed on schedule. - Regulation (EC) 1831/2003, Art. 11(2): antibiotics other than coccidiostats and histomonostats “may be marketed and used as feed additives only until 31 December 2005; as from 1 January 2006, those substances shall be deleted from the Register”. - Recital 26 frames the date as allowing “sufficient time for the development of alternative products”.

The litigation upheld the precautionary withdrawal. - Pfizer Animal Health v Council (Case T-13/99, Court of First Instance, judgment of 11 September 2002) dismissed Pfizer’s challenge to the 1998 withdrawal of virginiamycin and ordered Pfizer to pay the Council’s costs. The companion case Alpharma (T-70/99) was decided the same day (EPRS 2015, p. 10). - The court held that where there is scientific uncertainty, institutions “may, by reason of the precautionary principle, take protective measures without having to wait until the reality and seriousness of those risks become fully apparent” (para. 139). - It also held that “a preventive measure cannot properly be based on a purely hypothetical approach to the risk, founded on mere conjecture which has not been scientifically verified” (para. 143). - The link between growth promoters and human resistance “had not yet been scientifically proved but was nevertheless corroborated by a certain amount of reliable scientific data”. It was therefore for the Council “to exercise its discretion and assume its political responsibilities” (para. 443). The institutions “cannot be criticised” for withdrawing the authorisation provisionally while research continued (para. 444). - This was so even though the EU’s Scientific Committee on Animal Nutrition (SCAN) had concluded that virginiamycin use “does not constitute an immediate risk to public health in Denmark” (quoted at para. 214).

Later regulation went further. - EU veterinary medicines regulation. Regulation (EU) 2019/6, applicable from 28 January 2022, provides that antimicrobials “shall not be used in animals for the purpose of promoting growth nor to increase yield” (Art. 107(2)). They “shall not be applied routinely nor used to compensate for poor hygiene” (Art. 107(1)), and prophylaxis is confined to “exceptional cases” (Art. 107(3)). - Imports. Art. 118(1) extends the growth-promotion ban to operators in third countries exporting to the EU. Its implementing rules apply from 3 September 2026 (Delegated Reg. (EU) 2023/905; Implementing Reg. (EU) 2024/2598, “It shall apply from 3 September 2026”). - WHO. On 7 November 2017 the WHO recommended that farmers and the food industry stop routine use of medically important antibiotics for growth promotion and disease prevention in healthy animals (WHO 2017 guidelines page). - United States. FDA completed implementation of its Guidance for Industry #213 in January 2017. This removed production (growth-promotion) indications for medically important antimicrobials and brought the rest under veterinary oversight (FDA timeline).

Evidence on effects. - Systematic review. Tang et al. (2017, Lancet Planet Health 1:e316–e327) pooled 181 studies. Restrictions were “associated with a reduction in the presence of antibiotic-resistant bacteria” in animals, with pooled reductions of about 10–15% depending on antibiotic class. In human groups, resistance was “24% lower in the intervention groups”, most strongly for people in direct contact with food animals. The implications for the general population “remain unclear”. - EU surveillance. Between 2014 and 2021, total antimicrobial consumption in food-producing animals in the EU/EEA fell by 44%, while human use “remained relatively stable”. By 2021 it was lower per kg of biomass than human consumption (92.6 vs 125.0 mg/kg). The agencies found that “for certain combinations of bacteria and antimicrobials, AMR in bacteria from humans was associated with AMR in bacteria from food-producing animals which, in turn, was related to AMC in animals”. They concluded that measures to reduce consumption “have been effective in many countries” (JIACRA IV, EFSA Journal 22(2):e8589, February 2024). - A natural experiment on “loud and late” warnings. - Plasmid-mediated colistin resistance (mcr-1) was discovered in Chinese food animals in 2015, heralding “the breach of the last group of antibiotics” (Liu et al. 2016). - After China banned colistin as a growth promoter on 30 April 2017, colistin-resistant E. coli in pig faeces fell from 34.0% (2015–16) to 5.1% (2017–18). Human carriage of mcr-1-positive E. coli fell from 14.3% (2016) to 6.3% (2019) (Wang et al. 2020, Lancet Infect Dis 20:1161–1171). - This is the clearest post-2001 evidence that removing an agricultural use can reduce resistance in humans within a few years.

Evidence that complicates the picture. - Unintended consequences. Three years after the late-1990s EU bans, Casewell et al. (2003, J Antimicrob Chemother 52:159–161) reported that the main human benefit was lower resistance in enterococci. They also documented worse animal health and a rise in therapeutic use of tetracyclines, aminoglycosides, macrolides and other drugs “crucial in human medicine”, and urged reconsideration. - The longer Danish record. Danish pig data show consumption fell from 100 mg/kg of pig produced (1992) to 31 mg/kg (1999), then rose again to 49 mg/kg (2008), while productivity improved (Aarestrup et al. 2010, Am J Vet Res 71:726–733). The substitution effect was real but partial. Total use stayed about half its pre-ban level. - Scale of the wider problem. The Global Burden of Disease (GBD) antimicrobial resistance collaborators (2024) estimate 1.14 million deaths attributable to bacterial AMR in 2021, forecast to reach 1.91 million by 2050. Their estimates do not separate the agricultural contribution, which remains debated. The farm-to-human share of the human burden is still not quantified with confidence.

Verdict: strengthened. The specific precautionary step the chapter endorsed was implemented on schedule. It survived judicial challenge, was extended and generalised internationally, and has measurable resistance benefits in animals, in exposed humans, and in one large natural experiment in the general population. The size of the general-population health benefit remains uncertain, and the early risk-for-risk trade-off was real.

Implications for weight. This is the chapter’s strongest worked example, and it supports a transferable mechanism: act on a plausible pathway to serious, hard-to-reverse harm before the size of that harm is known, when the benefit forgone is modest and alternatives exist. The Pfizer judgment shows the legal form this can take: action on “reliable” but incomplete data, not on “mere conjecture”, with provisional status and continued research. The case also shows two further lessons: - The costs of precaution fall partly on the precautionary side itself (more therapeutic use), so they must be monitored. - Knowledge can move from “uncertainty” towards “risk” for some sub-questions and not others.

Claim 5: The level of proof is a political choice that shifts the costs of being wrong, and should depend on harm, benefits, alternatives and error costs both ways (p. 193)#

Original claim (p. 193). “Levels of proof” range from the Commission’s “reasonable grounds for concern” to “beyond reasonable doubt”. Choosing among them “can radically shift the size, nature and distribution of the costs of being wrong. This is a key political decision with profound ethical implications.” The appropriate level “depends upon the size and nature of the potential harm, the claimed benefits, the available alternatives, and the potential costs of being wrong in both directions”.

Subsequent developments

Courts confirmed the political character of the choice. - The Pfizer court held that the Commission and Council “may disregard the conclusions drawn in the SCAN opinion” (para. 200). The reason given was “the political responsibilities and democratic legitimacy of the Commission”: the members of SCAN, “although they have scientific legitimacy, have neither democratic legitimacy nor political responsibilities” (para. 201). - It also set a lower bound: not “purely hypothetical” risk, but risk “adequately backed up by the scientific data available at the time” (para. 143). - This is close to the chapter’s position that science informs but does not determine the evidential threshold.

Some regimes now tie graded strengths of evidence to different consequences. - EU pesticides. Approval is barred for substances that are or should be classified “as mutagen category 1A or 1B”, “carcinogen category 1A or 1B” or “toxic for reproduction category 1A or 1B”, with exposure-based exceptions for the last two. It is also barred for endocrine disruptors “that may cause adverse effect in humans” (Reg. (EC) 1107/2009, Annex II, points 3.6.2–3.6.5). - The EU classification regulation grades evidence into categories 1A, 1B and 2 (roughly known, presumed and suspected; Reg. (EC) 1272/2008, Annex I, not re-retrieved here). - This is an explicit legislative choice to act on “presumed” as well as “known” hazard, and to set aside case-by-case risk-benefit balancing for those categories. - US chemicals law. The 2016 amendments to the Toxic Substances Control Act require EPA to decide whether a chemical presents an unreasonable risk “without consideration of costs or other nonrisk factors”, including risks to “potentially exposed or susceptible subpopulations”. Once such a risk is found, EPA must regulate “so that the chemical substance or mixture no longer presents such risk” (15 U.S.C. § 2605(b)(4)(A), (a)). This is a statutory choice about who bears the cost of error at the risk-finding stage. - The EEA itself. The 2013 volume put “the use of different strengths of evidence for different purposes” at the centre of its concluding chapter (Gee, Ch 27; EEA 2013 summary, p. 37). Its working definition calls for “an appropriate strength of scientific evidence” (EEA 2013, p. 649, via EPRS 2015, p. 10). - UNESCO. COMEST’s 2005 working definition sets the threshold at harm that is “scientifically plausible but uncertain” (EPRS 2015, p. 10).

But the explicit weighing the chapter recommends is rare, and disputes persist. - Glyphosate. - IARC classified glyphosate as “probably carcinogenic to humans” (Group 2A) in 2015 (IARC Monographs vol. 112; not re-retrieved here). - ECHA’s Committee for Risk Assessment concluded in May 2022 that glyphosate “does not meet the criteria to be classified as carcinogenic, mutagenic or toxic for reproduction”. - When renewal came to a vote, the member-state Standing Committee “has not delivered an opinion” and the Appeal Committee “did not deliver an opinion” either. The Commission renewed approval until 15 December 2033 (Implementing Reg. (EU) 2023/2660, recitals and Annex). - Different bodies applied different evidential questions (hazard identification versus classification criteria), and the political bodies declined to own the choice. - Explicit two-way weighing. I found no regime that routinely states the chapter’s full set of factors (harm, benefits, alternatives, error costs both ways) when fixing an evidence threshold. The EPRS reports that institutions “do not agree on the method for determining when to apply precautionary measures (cost-benefit analysis, risk trade-off analysis, cost-effectiveness analysis, pros and cons analysis of action and inaction, etc.)” (EPRS 2015, p. 1).

Verdict: strengthened as an analytical claim: courts and legislatures have openly treated evidence thresholds as political, graded choices. Partly held up as a prescription: explicit, case-by-case weighing of error costs in both directions is still uncommon, and some regimes have moved in opposite directions (EU hazard cut-offs; US exclusion of costs from risk findings).

Implications for weight. This is the chapter’s most durable conceptual contribution, and it is technology-neutral. Whoever sets the evidence threshold decides who bears the cost of being wrong, so the threshold should be chosen openly with reference to the stakes on both sides. As a lens, it directs attention to three questions: - Where, in any governance system, is the threshold set? - Is the choice made explicitly or by default, as with glyphosate? - Who bears the cost if the threshold is too high or too low?

Claim 6: Early, wide stakeholder involvement adds information and may improve trust without stifling innovation or compromising science (p. 193)#

Original claim (p. 193). Involving “a wide range of stakeholders … at the earliest stage of the appraisal and choice of technological and social options” brings “two key benefits”. It “augments the information available to policy-making, but may also improve public trust … without necessarily stifling innovation or compromising science”.

Subsequent developments

Policy uptake. - The EU’s 7th Environment Action Programme (2013) stated that “major technological innovations should be accompanied by public dialogues and participatory processes” (as quoted in EPRS 2015, p. 19). - Horizon Europe commits to “engage and involve all societal actors, such as citizens and civil society organisations, in co-designing and co-creating responsible research and innovation (RRI) agendas” (Reg. (EU) 2021/695, recitals). - Owen, von Schomberg & Macnaghten (2021) trace how RRI emerged in the Commission and was operationalised in Horizon 2020 through five “keys”, public engagement among them. They reflect on its prospects under Horizon Europe.

Evidence from the flagship participatory exercise on a technology in the report. - The UK’s GM Nation? public debate (2003) was, per its official independent evaluators, “an unprecedented experiment for the UK in public engagement”. It “suffered from a number of important flaws: in terms of both design and implementation”, which “in part reflect constraints on time and budget, and the context of political pressure” (Horlick-Jones et al. 2006, New Genetics and Society 25:265–288). - Its 36,000 open questionnaire responses “were indeed not fully representative of British ‘public opinion’”. A representative survey found opinion “fragmented, with considerable ambivalence coexisting alongside outright opposition” (Pidgeon et al. 2005, Risk Analysis 25:467–479). - The exercise did add information, notably the heterogeneity of opinion. It did not produce a settled or trusted outcome.

Evidence on trust and legitimacy. - A preregistered survey experiment in Ireland (N = 1,309) found that deliberative mini-publics “increase legitimacy perceptions among the broader citizenry; however, these beneficial effects are largely limited to situations in which their recommendations are honored” (Germann et al. 2022, Political Studies 72:677–700). - Public support for citizens’ assemblies across 15 Western European countries is partly “outcome contingent” (Pilet et al. 2022, Eur J Polit Res 62:873–902). - Interviews with environmental-governance practitioners found “signs that the participatory agenda has started to lose its momentum and justification because of disappointments about actual achievements” (Wesselink et al. 2011, Environment and Planning A 43:2688–2704).

Evidence on innovation. - Tait (2009, EMBO Reports 10:S18–S22) argued from the European GM experience that upstream engagement had been dominated by activist interests. She pointed to destroyed crop trials and advice that “could steer potential applicants away from areas … that are known to be contentious”. Rather than abandoning engagement, she proposed “rules for engagement” setting standards for evidence and listening. - This is an opinion piece, not an empirical test. I found no empirical study that tests whether early stakeholder involvement leaves innovation rates unaffected.

Verdict: partly held up. - The information benefit is supported: engagement exercises surfaced views and framings that expert processes missed. - The trust benefit is conditional. It appears mainly when participation visibly shapes decisions, and can backfire when it does not. - The claim that engagement need not stifle innovation is untested, and contested by commentators drawing on the GM case.

Implications for weight. Use the information claim. Treat the trust claim as conditional on uptake: participation that is ignored may lower trust. Treat the innovation claim as an open question. For a lens, the useful mechanism is that the legitimacy effect of participation depends on whether it changes outcomes, not on whether it happens.

Claim 7: Recent controversies (GMOs, oil-rig disposal) turn on public values and uncertainties, not demands for unequivocal proof; recognising this is a first step to better governance (p. 193)#

Original claim (p. 193). “Recent controversies over emerging technologies such as genetically modified organisms and oil-rig disposal have much to do with public values and scientific uncertainties, in contrast to the previously low emphasis on values and to the demand for unequivocal scientific proof before action.” Recognising this “is a first step towards their improved governance.”

Subsequent developments

Brent Spar (the “oil-rig disposal” case). - Facts that complicate the chapter’s reading. - Brent Spar was not an oil rig but “a floating oil tank and never a fixed installation” (House of Lords, 15 October 1996, Hansard) or “floating storage buoy” (Commons, 13 May 1998). - Greenpeace occupied it from 30 April 1995. Shell abandoned deep-sea disposal on 20 June 1995. - Greenpeace admitted on 5 September 1995 that its claim that the Spar held 5,500 tonnes of oil was inaccurate, and apologised to Shell. An audit by the classification society DNV, published 18 October 1995, supported Shell’s estimate. These dates come from the Wikipedia article (secondary). The parliamentary record corroborates the substance: a Lords speaker referred in 1996 to “an unfortunate inaccuracy in a statement by Greenpeace”. In the Commons on 24 October 1995, Mr Robertson said the DNV study “demonstrates that Shell had been right all along”. - The science. - The group convened by the Natural Environment Research Council (NERC) under Professor John Shepherd reported on 22 May 1996. It found that the “global impacts on the environment and on human health of the deep-sea disposal of a structure such as the ‘Brent Spar’ would be very small—roughly equivalent to … the wreck in the deep ocean of a ship of 30,000 to 50,000 tonnes”. Those impacts were “not large enough to be a crucial factor in the selection of the best disposal options” (as reported to the House of Lords by the Government spokesperson, Baroness Miller of Hendon, 15 October 1996). - The same debate heard that Shell’s option appraisal was “a fairly poor example” of the “best practicable environmental option” (BPEO) process (Earl of Selborne). - So the scientific uncertainty was modest. What drove the outcome was a value judgement against using the sea for disposal, amplified by a factual error. - The outcome. - Shell announced reuse of the Spar as a quay at Mekjarvik, Norway, on 29 January 1998 (Wikipedia; secondary). - OSPAR Decision 98/3 (July 1998) prohibits “the dumping, and leaving wholly or partly in place, of disused offshore installations”. It allows derogations for steel jackets over 10,000 tonnes and for concrete gravity-based structures. OSPAR reports about ten derogations granted since (OSPAR, “Offshore installations” page). - That is, the value judgement was codified, with an explicit, case-by-case exception process. - The Brent field’s own decommissioning shows the value conflict persists. Shell’s programme to leave the Brent Bravo, Charlie and Delta concrete gravity-based structures in place was still listed among “draft decommissioning programmes under consideration” on the UK regulator OPRED’s page as last updated on 25 September 2026.

GMOs. - The science largely converged. The US National Academies (2016) found “no substantiated evidence that foods from GE crops were less safe”, no adverse effects in long-term livestock data, and “sweeping statements about GE crops are problematic because issues related to them are multidimensional”. - Governance moved explicitly to values. - The Academies observed that national regulatory processes “vary greatly because they mirror the broader social, political, legal, and cultural differences among countries”. - The EU then made values a formal legal ground. Directive (EU) 2015/412 lets member states restrict cultivation of an EU-authorised GMO on grounds “distinct from and complementary to” the EFSA-led risk assessment. These include “environmental or agricultural policy objectives”, “town and country planning, land use, socioeconomic impacts, coexistence and public policy”, and “cultural traditions” (recitals 13–15; Art. 26b(3), which lists seven grounds and bars “public policy” from being invoked on its own). - The Court of Justice held that organisms from newer mutagenesis techniques are GMOs under the 2001 Directive (C-528/16, 25 July 2018). - The EU adopted a separate regulation for plants from new genomic techniques, signed 17 June 2026 and published 26 June 2026 (Legislative Observatory, procedure 2023/0226(COD)). I have not checked the final text’s contents. - Two decades of controversy therefore produced recognition of values in law. They did not produce a settled or widely trusted outcome.

Verdict: partly held up. - The central claim, that such controversies turn on values and that governance must say so openly, is strongly supported by the EU’s 2015 turn to non-scientific grounds for GMO decisions and by the OSPAR codification after Brent Spar. - The chapter’s pairing of “values and scientific uncertainties” fits the GMO case less well after 2016, as health-risk assessments converged. - It fits Brent Spar poorly. There the science suggested small impacts, and a factual error by a campaigner shaped the outcome. - “Open recognition” happened. Improved governance, in the sense of less conflict, did not clearly follow.

Implications for weight. The lens holds: controversies over technologies are often about which harms and benefits count and who decides, not about missing data. Two refinements from hindsight should accompany it: - In a values conflict, factual claims by any party still need checking. Brent Spar shows a value-driven outcome can rest partly on an error. - Formally recognising values (opt-outs, derogation procedures) can channel conflict without resolving it.

Claim 8: The precautionary principle stimulates more innovation (through diversity and flexibility) and better science (p. 194)#

Original claim (p. 194). “The use of the precautionary principle can also bring benefits beyond the reduction of health and environmental impacts, stimulating both more innovation, via technological diversity and flexibility, and better science.”

Subsequent developments

The regulation-and-innovation literature (the nearest empirical test). - Ambec, Cohen, Elgie & Lanoie (“The Porter Hypothesis at 20”, RFF discussion paper 2011; Review of Environmental Economics and Policy 7(1):2–22, 2013) conclude that evidence for the “weak” version (“stricter regulation leads to more innovation”) is “fairly well established”. Evidence for the “strong” version (“stricter regulation enhances business performance”) “is mixed, with more recent studies providing more supportive results” (RFF DP 11-01, p. 16). - Cohen & Tubb’s meta-analysis of 103 publications and over 2,000 effect sizes (JAERE 5(2):371–399, 2018) found “considerable heterogeneity in both the sign and significance”. A positive effect is “more likely at the state, region, or country level” than at firm level, “although in both cases the most likely scenario is statistical insignificance”. - Dechezleprêtre & Sato (REEP 11(2):183–206, 2017) find that environmental regulations induce “innovation in clean technologies” but that “the resulting benefits do not appear to be large enough” to offset compliance costs for regulated firms. Short-run adverse effects on trade, employment and productivity are “small relative to general trends”. - None of these tests precautionary regulation specifically, meaning action under uncertainty. They test environmental regulation generally.

Precaution-specific arguments and cases. - Todt & Luján (2014, Risk Analysis 34:2163) argue conceptually that precautionary regulation “does not necessarily curtail scientific-technological innovation” and tends “to imply innovation in methods, products, and processes”. This is an argument, not a measurement. - The ozone case shows induced innovation with a sting: the chosen substitutes (HFCs) became a climate problem requiring the Kigali Amendment (2016). - The BPA ban (2024) relies on industry being “proactive in preparing for a transition” but grants 18- and 36-month transitions because alternatives for some can linings are not yet at “commercial scale” (Reg. (EU) 2024/3190, recitals). - The antibiotic growth-promoter ban was paired with time for “development of alternative products” (Reg. 1831/2003, recital 26). The Danish data show productivity rising while antimicrobial use fell (Aarestrup et al. 2010), which is weak evidence of adaptive process innovation. - GM crops. The 2013 EEA volume argued that GM-focused intellectual property “often close[s] down, rather than open[s] up further innovation potential” and favoured agroecology (EEA 2013 summary, p. 29). Tait (2009) argued the reverse: the European GM experience deterred research in contentious areas. Neither side offers systematic measurement.

The claim became politically contested. - Industry associations drafted an “innovation principle” to sit alongside precaution (European Risk Forum, BusinessEurope and the European Round Table of Industrialists, 2015; EPRS 2015, p. 22). - Garnett, Van Calster & Reins (2018, Law, Innovation and Technology 10:1–14) describe this as “the first time an industry association has successfully tried to introduce a new principle into the EU’s legal order”. - Horizon Europe’s founding regulation now says activities should “take advantage of and inspire innovation-friendly regulation, in line with the innovation principle” (Reg. (EU) 2021/695, recital 6). The regulation does not mention the precautionary principle. - The 2013 EEA volume asserted that “there is growing evidence that precautionary measures do not stifle innovation, but instead can encourage it, in particular when supported by smart regulation or well-designed tax changes” (EEA 2013 summary, p. 38). It is the same organisation restating its own thesis, not independent confirmation.

Verdict: contested. - A weak form is supported: regulation, including precautionary regulation, redirects innovation towards substitutes and cleaner processes. - The stronger reading, that precaution produces more innovation overall or better outcomes for regulated firms, is not supported by the meta-analytic evidence, which centres on no significant effect. - Induced substitutes have sometimes created new hazards. - The “better science” half of the claim was not tested in any source I found.

Implications for weight. Cite this as “precaution can redirect innovation”, with the substitution caveat, not “precaution stimulates more innovation”. The better-supported transferable mechanism is that deadlines and restrictions create demand for alternatives, whose own hazards then need appraising (lesson 7). The emergence of a rival “innovation principle” shows that the balance between the two is fought over politically, as the chapter’s own closing sentence anticipates.

Claim 9: Over-precaution can be expensive (lost innovation, lost lines of inquiry), but the lessons would improve the balance (p. 194)#

Original claim (p. 194). The lessons “may also help to achieve a better balance between proportionate and precautionary public policies, recognising that over-precaution can also be expensive, in terms of lost opportunities for innovation and lost lines of scientific inquiry.”

Subsequent developments

How frequent are false positives? - Hansen, Krayer von Krauss & Tickner (2007, Risk Analysis 27:255–269) examined 88 alleged false positives. Most turned out to be “real risks”, “‘The jury is still out’”, “nonregulated proclaimed risks”, “‘Too narrow a definition of risk’” or “risk-risk tradeoffs”. Only four qualified as genuine regulatory false positives: southern corn leaf blight, swine flu, saccharin, and food irradiation (for consumer health). They concluded that “concerns about false positives do not represent a reasonable argument against future application of the precautionary principle”. - The EEA reprinted the analysis as Ch 2 of the 2013 volume. It concluded that “fear of false positives is misplaced”, that costs were “mainly economic” except swine flu (which caused “some unintended deaths”), and that “determining the net costs of mistaken regulatory action … requires a complete assessment” (EEA 2013 summary, p. 12). - Cox (2007, Risk Analysis 27:1083–1086) argued the classification “applies a diverse set of questionable criteria to label many highly uncertain risks as ‘real’ even when no real or potential harm has actually been demonstrated”. Examples include treating conservative regulatory assumptions “as if they were true values” and ambiguous associations “as if they were known to be true causal relations”. Such criteria “can classify even nonexistent and unknown risks as ‘real’”. - Two structural limits: - By design the count excludes alarms that were never regulated. For example, the 2013 volume’s mobile-phone warning has not been borne out (Karipidis et al. 2024), but it was never regulated, so it would not count. - The count does not capture the cost of forgone options that never reached a regulator.

Claimed costs of over-precaution. - Wesseler & Zilberman (2014, Environment and Development Economics 19:724–742) use a real-options model to estimate that delay in approving vitamin-A-enriched (“Golden”) rice in India implied “about 1.4 million life years lost over the past decade”. This is a model-based estimate, contingent on assumptions and contested, but it shows the form such costs take. - Tait (2009) documents lost lines of inquiry in European plant biotechnology, including destroyed field trials. This is descriptive, not quantified.

Risk-for-risk trade-offs in the section’s own domain. - Antibiotic growth promoters. Therapeutic use of medically important classes rose after the bans (Casewell et al. 2003), although total use stayed about half its pre-ban level in Denmark (Aarestrup et al. 2010). - CFCs to HFCs. Kigali Amendment, 2016. - PBDEs to organophosphate esters. Qadeer et al. 2024. - DDT. The Stockholm Convention builds a risk-risk trade-off into its text: DDT is in Annex B “with acceptable purpose for disease vector control” (pops.int). That is an explicit judgement that malaria risk can outweigh persistent-pollutant risk. - The EPRS notes the classic example of nitrite preservatives in meat, which carry a carcinogen concern but prevent potentially fatal food poisoning (EPRS 2015, p. 20). - These trade-offs were argued before 2001 (Graham & Wiener, Risk vs. Risk, 1995) and developed after it (Sunstein, Laws of Fear, 2005). Ch 17 does not engage them.

Do the lessons improve the balance? I found no evaluation of that. Lessons 6 and 7 (scrutinise benefits; evaluate alternatives) are the ones that would catch risk-for-risk problems. The regrettable-substitution cases suggest they are applied unevenly. The EU’s 2024 BPA ban, which extends controls to classified bisphenol analogues, is one instance where they were applied.

Verdict: partly held up. - The concession that over-precaution has costs was correct. Later work documents examples and at least one large, contested cost estimate. - The implied empirical premise, that false positives are rare, rests mainly on one analysis by authors close to the reports, whose criteria were disputed. - Risk-for-risk trade-offs turned out to be common in the section’s own flagship cases. - That the lessons “would improve the balance” is untested.

Implications for weight. Use the section’s symmetric framing of error costs, but do not rely on “false positives are rare” as settled. The better-grounded point is that the costs of precaution often arise through substitution and displacement, not through outright false alarms. Appraisal should look for what fills the gap when something is restricted.

Claim 10: Maximising innovation while minimising hazard is political, and would be more successful if it embraced the twelve lessons (p. 194, central conclusion)#

Original claim (p. 194). “It is the central conclusion of this report that the very difficult task of maximising innovation whilst minimising hazards to people and their environments, which is ultimately a matter of political discourse, could be more successful if it embraced the twelve late lessons”.

Subsequent developments

No institution adopted the twelve lessons as a package, and there is no comparative evaluation. - The lessons were restated unchanged in the 2013 volume (summary box, p. 11) and summarised by the European Parliament’s research service (EPRS 2015). - I found no regulator that adopted them as operating rules, and no study comparing the “surprise” record of institutions that did and did not follow them.

The EEA’s own twelve-year retrospective (2013) was mixed. - Progress: “the number of stakeholders involved in decision-making has become larger and more diverse”, and there was “increasing attention to communicating scientific uncertainty”. - But “many of the political and scientific ‘bureaucratic silos’ still remain”, with “most polluters still not paying the full costs of pollution” (EEA 2013 summary, p. 38). - Gee’s concluding chapter: “The capacity to foresee and forestall disasters, especially when such action is opposed by powerful economic and political interests, appears to be limited” (p. 37).

Comparative evidence on “more precautionary” jurisdictions. - Hammitt et al. (2005, Risk Analysis 25:1215–1228) coded relative US–EU precaution for 100 randomly sampled risks from almost 3,000, for each year 1970–2004. Averaging over risks, they found “no significant difference in relative precaution over the period”, “a modest shift toward greater relative precaution of European regulation since about 1990”, and “a diversity of trends across risks”. - If even the EU/US contrast is issue-specific, “adopters” of precaution are hard to identify, which makes the chapter’s claim difficult to test.

Surprises continued in the jurisdiction most formally committed to precaution. - The EU has had the precautionary principle in its treaty since 1992 and in the Commission’s 2000 Communication. Yet since 2001 it has faced: - PFAS exposure above EFSA’s 2020 tolerable weekly intake for parts of the population - dietary BPA exposure exceeding EFSA’s 2023 tolerable daily intake by two to three orders of magnitude - two decades from first bee warnings to the 2018 neonicotinoid restrictions - These largely reflect exposures that began before 2001 and the slowness of action, not failure of the principle as such. They show that formal commitment did not avert costly surprises.

Where specific lessons were institutionalised, there is evidence of benefit. - Monitoring (lesson 2) detected illegal CFC-11 emissions (Montzka et al. 2018) and tracked falling POP levels (Stockholm effectiveness evaluation 2023). - Acting on reasonable grounds for concern (lesson 12) in antibiotic growth promoters and colistin produced measurable reductions in resistance (JIACRA IV 2024; Wang et al. 2020). - Independent assessment (lesson 10). The 2013 volume identified “the value of independent scientific research and risk assessments” as a common theme. The PFAS industry-documents analysis (Gaber et al. 2023) is consistent with its absence being costly. - These are lesson-by-lesson successes, not evidence that the package produces fewer or cheaper surprises overall.

Verdict: unclear. The claim is a counterfactual about an intervention (embracing all twelve lessons) that no institution made, so it cannot be tested directly. Individual lessons have good post-2001 support: monitoring, acting on reasonable grounds, independence, and appraising alternatives. Formal commitment to precaution without them did not prevent costly surprises.

Implications for weight. Use the twelve lessons as a checklist of failure modes, each carrying its own level of evidence (see Claims 1–9). Do not use them as a validated package. The “ultimately a matter of political discourse” framing holds up well. After 2001 the decisive moves were political and legal: the court’s Pfizer reasoning, OSPAR’s codification, the EU’s GMO opt-out, the glyphosate “no opinion” votes, and the “innovation principle” campaign. That supports reading the chapter chiefly as an argument about where decisions are made, not as a technical recipe.

Summary of verdicts#

Technology-neutral lessons this check supports (for later use as a lens)#

Each is tied to the chapter’s pages and to the later evidence above.

1. The evidence threshold distributes the cost of error. Whoever sets it, and whether they do so openly or by default, determines who pays for being wrong (p. 193; Pfizer paras 143, 200–201; glyphosate renewal).
2. Screen on properties that predict the scale and irreversibility of harm when the specific harm is unknown, and pair screening with monitoring. Examples of such properties are persistence, accumulation, mobility and reach (Table 17.1, p. 192; Stockholm evaluation 2023; Montzka et al. 2018; Cousins et al. 2019). Expect substitution towards unscreened analogues.
3. The endpoint that matters often lies outside the original appraisal frame (p. 192; BPA and PFAS immune endpoints, EFSA 2020, 2023).
4. Restriction creates substitutes whose hazards need appraising in turn (p. 192, “alternatives”; lesson 7; Kigali 2016; Qadeer et al. 2024; Reg. 2024/3190).
5. Provisional action on reliable but incomplete evidence, with continued research, is legally and practically workable, and can show benefits within years (Table 17.1; T-13/99 para. 444; Wang et al. 2020).
6. Participation raises legitimacy mainly when it visibly changes outcomes (p. 193; Germann et al. 2022; GM Nation? evaluations).
7. Formally recognising values can channel a conflict without resolving it, and value-driven outcomes can still rest partly on factual errors (p. 193; Directive 2015/412; Brent Spar).
8. Vindicated warnings do not give a base rate. Some prominent early warnings do not hold up, and a record built from selected cases cannot say which will (pp. 192, 194; Karipidis et al. 2024; Hansen et al. 2007 versus Cox 2007).

Sources#

All retrieved 25 September 2026 unless noted.

EEA and EU institutional - EEA (2001). Late lessons from early warnings: the precautionary principle 1896–2000. Environmental Issue Report No 22. Ch 17, pp. 192–194 (project extract). - EEA (2013). Late lessons from early warnings: science, precaution, innovation — Summary (EEA Report 1/2013). Publication page https://www.eea.europa.eu/en/analysis/publications/late-lessons-2 (summary PDF served at https://www.eea.europa.eu/en/analysis/publications/late-lessons-2/late-lessons-2/@@download/file). Pages cited: 9–12, 29, 31, 37–38. - EEA (2019). Emerging chemical risks in Europe — ‘PFAS’. Briefing 12/2019, 16 December 2019. https://www.eea.europa.eu/en/analysis/publications/emerging-chemical-risks-in-europe - Bourguignon, D. (2015). The precautionary principle: definitions, applications and governance. EPRS In-depth analysis PE 573.876, December 2015. https://www.europarl.europa.eu/RegData/etudes/IDAN/2015/573876/EPRS_IDA(2015)573876_EN.pdf - European Parliament Legislative Observatory, procedure 2023/0226(COD) (plants obtained by certain new genomic techniques): final act signed 17 June 2026, published in the Official Journal 26 June 2026. https://oeil.secure.europarl.europa.eu/oeil/en/procedure-file?reference=2023/0226(COD)

EU legislation and case law (EUR-Lex) - Case T-13/99 Pfizer Animal Health SA v Council, Court of First Instance, judgment of 11 September 2002, paras 119, 139, 143, 200–201, 214, 443–444, operative part. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:61999TJ0013 - Regulation (EC) No 1831/2003 on additives for use in animal nutrition, Art. 11(2), recitals 25–26. https://eur-lex.europa.eu/eli/reg/2003/1831/oj - Regulation (EU) 2019/6 on veterinary medicinal products, Arts 107, 118, 160. https://eur-lex.europa.eu/eli/reg/2019/6/oj - Commission Delegated Regulation (EU) 2023/905 (27 February 2023) on the import prohibition under Art. 118. https://eur-lex.europa.eu/eli/reg_del/2023/905/oj - Commission Implementing Regulation (EU) 2024/2598 (4 October 2024), applying from 3 September 2026. https://eur-lex.europa.eu/eli/reg_impl/2024/2598/oj - Regulation (EC) No 1107/2009 on plant protection products, Annex II points 3.6.2–3.6.5, 3.7.1–3.7.2. https://eur-lex.europa.eu/eli/reg/2009/1107/oj - Commission Delegated Regulation (EU) 2023/707 (new CLP hazard classes: endocrine disruption, PBT/vPvB, PMT/vPvM). https://eur-lex.europa.eu/eli/reg_del/2023/707/oj - Regulation (EC) No 1272/2008 (CLP), Annex I categories 1A/1B/2 (text not re-retrieved in this pass). https://eur-lex.europa.eu/eli/reg/2008/1272/oj - Commission Regulation (EU) 2024/3190 (19 December 2024) on BPA and other bisphenols in food-contact materials. https://eur-lex.europa.eu/eli/reg/2024/3190/oj - Commission Implementing Regulation (EU) 2018/783 (29 May 2018) on imidacloprid. https://eur-lex.europa.eu/eli/reg_impl/2018/783/oj - Case C-162/21, Court of Justice, judgment of 19 January 2023 (emergency authorisations for neonicotinoid-treated seeds). https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:62021CJ0162 - Commission Implementing Regulation (EU) 2023/2660 (28 November 2023) renewing glyphosate approval to 15 December 2033. https://eur-lex.europa.eu/eli/reg_impl/2023/2660/oj - Directive (EU) 2015/412 (11 March 2015) on member-state restriction of GMO cultivation. https://eur-lex.europa.eu/eli/dir/2015/412/oj - Case C-528/16 Confédération paysanne and Others, Court of Justice (Grand Chamber), judgment of 25 July 2018. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:62016CJ0528 - Regulation (EU) 2021/695 establishing Horizon Europe, recital 6 (innovation principle) and RRI recital. https://eur-lex.europa.eu/eli/reg/2021/695/oj

Other regulators, treaties and official bodies - Stockholm Convention Secretariat. “All POPs listed in the Stockholm Convention”. https://www.pops.int/TheConvention/ThePOPs/AllPOPs/tabid/2509/Default.aspx ; “The New POPs” (listing decisions 2009–2025). https://www.pops.int/TheConvention/ThePOPs/TheNewPOPs/tabid/2511/Default.aspx ; “Overview”. https://www.pops.int/TheConvention/Overview/tabid/3351/Default.aspx ; “Outcomes of the effectiveness evaluation of the Stockholm Convention (2023)”. https://www.pops.int/Implementation/EffectivenessEvaluation/Overview/tabid/5559/Default.aspx - UNEP Ozone Secretariat. Kigali Amendment (agreed 15 October 2016; in force 1 January 2019). https://ozone.unep.org/treaties/montreal-protocol/amendments/kigali-amendment-2016-amendment-montreal-protocol-agreed - WHO (2017). WHO guidelines on use of medically important antimicrobials in food-producing animals, 7 November 2017. https://www.who.int/publications/i/item/9789241550130 - US FDA. “Timeline of FDA Action on Antimicrobial Resistance” (GFI #209, #213; implementation completed January 2017). https://www.fda.gov/animal-veterinary/antimicrobial-resistance/timeline-fda-action-antimicrobial-resistance - 15 U.S.C. § 2605 (TSCA § 6, as amended 2016). https://www.law.cornell.edu/uscode/text/15/2605 - ECDC/EFSA/EMA (2024). JIACRA IV — 2019–2021. EFSA Journal 22(2):e8589, February 2024. https://doi.org/10.2903/j.efsa.2024.8589 - EFSA CONTAM Panel (2020). Risk to human health related to the presence of perfluoroalkyl substances in food. EFSA Journal 18:e06223. https://doi.org/10.2903/j.efsa.2020.6223 - EFSA CEP Panel (2023). Re-evaluation of the risks to public health related to the presence of bisphenol A (BPA) in foodstuffs. EFSA Journal 21:e06857. https://doi.org/10.2903/j.efsa.2023.6857 - IARC (2015/2017). Monographs vol. 112 (glyphosate, Group 2A). https://publications.iarc.who.int/549 (not re-retrieved in this pass) - National Academies of Sciences, Engineering, and Medicine (2016). Genetically Engineered Crops: Experiences and Prospects, released 17 May 2016. Executive Summary. https://nap.nationalacademies.org/read/23395/chapter/2 ; https://doi.org/10.17226/23395 - IRGC. Risk governance framework (original white paper 2005, ed. O. Renn). https://irgc.org/risk-governance/irgc-risk-governance-framework/ - OSPAR Commission. “Offshore installations” (Decision 98/3; derogations). https://www.ospar.org/work-areas/oic/installations - UK OPRED/GOV.UK. “Oil and gas: decommissioning of offshore installations and pipelines”, last updated 25 September 2026 (Brent gravity-based structures listed under draft programmes under consideration; OSPAR 98/3 summary). https://www.gov.uk/guidance/oil-and-gas-decommissioning-of-offshore-installations-and-pipelines - UK Hansard (retrieved via the Parliament Hansard API): - House of Lords, 15 October 1996, “Oil and Gas Installation Decommissioning: Select Committee Report” (Baroness Miller of Hendon on the NERC/Shepherd group report of 22 May 1996; Earl of Selborne; Lord Dixon-Smith). https://hansard.parliament.uk/Lords/1996-10-15/debates/9194a984-3a59-47b2-9d4d-9894fb8dbbfc/OilAndGasInstallationDecommissioningSelectCommitteeReport - House of Commons, 24 October 1995, “Environment (Scotland)” (Mr Robertson on the DNV study). https://hansard.parliament.uk/Commons/1995-10-24/debates/f49a8a2a-b70c-48c7-8359-48f633b7b391/EnvironmentScotland - House of Commons, 13 May 1998, “Oil and Gas Platforms” (Mr Anthony D. Wright). Debate ID 07df8cd2-7b96-432a-9b68-3c66d69ac181.

Peer-reviewed and scholarly - Aarestrup, F.M. et al. (2010). Changes in the use of antimicrobials and the effects on productivity of swine farms in Denmark. Am J Vet Res 71(7):726–733. https://doi.org/10.2460/ajvr.71.7.726 - Ambec, S., Cohen, M.A., Elgie, S., Lanoie, P. (2013). The Porter Hypothesis at 20. Review of Environmental Economics and Policy 7(1):2–22. https://doi.org/10.1093/reep/res016 ; discussion paper version RFF DP 11-01 (January 2011). https://media.rff.org/documents/RFF-DP-11-01.pdf - Aven, T. (2011). On different types of uncertainties in the context of the precautionary principle. Risk Analysis 31:1515–1525. https://doi.org/10.1111/j.1539-6924.2011.01612.x - Casewell, M. et al. (2003). The European ban on growth-promoting antibiotics and emerging consequences for human and animal health. J Antimicrob Chemother 52(2):159–161. https://doi.org/10.1093/jac/dkg313 - Cohen, M.A., Tubb, A. (2018). The impact of environmental regulation on firm and country competitiveness: a meta-analysis of the Porter hypothesis. JAERE 5(2):371–399. https://doi.org/10.1086/695613 (abstract via OpenAlex) - Cousins, I.T. et al. (2019). Why is high persistence alone a major cause of concern? Environ Sci: Processes Impacts 21:781–792. https://doi.org/10.1039/c8em00515j - Cousins, I.T. et al. (2020). The high persistence of PFAS is sufficient for their management as a chemical class. Environ Sci: Processes Impacts 22:2307–2312. https://doi.org/10.1039/d0em00355g - Cox, L.A. (2007). Regulatory false positives: true, false, or uncertain? Risk Analysis 27:1083–1086 (author reply 1087–1089). https://doi.org/10.1111/j.1539-6924.2007.00975.x - Dechezleprêtre, A., Sato, M. (2017). The impacts of environmental regulations on competitiveness. REEP 11(2):183–206. https://doi.org/10.1093/reep/rex013 - Gaber, N., Bero, L., Woodruff, T.J. (2023). The Devil they Knew: chemical documents analysis of industry influence on PFAS science. Annals of Global Health 89(1):37. https://doi.org/10.5334/aogh.4013 - Garnett, K., Van Calster, G., Reins, L. (2018). Towards an innovation principle: an industry trump or shortening the odds on environmental protection? Law, Innovation and Technology 10(1):1–14. https://doi.org/10.1080/17579961.2018.1455023 - GBD 2021 Antimicrobial Resistance Collaborators (2024). Global burden of bacterial antimicrobial resistance 1990–2021 with forecasts to 2050. Lancet 404(10459):1199–1226. https://doi.org/10.1016/S0140-6736(24)01867-1 - Germann, M., Mariën, S., Muradova, L. (2022/2024). Scaling up? Unpacking the effect of deliberative mini-publics on legitimacy perceptions. Political Studies 72:677–700. https://doi.org/10.1177/00323217221137444 - Giesy, J.P., Kannan, K. (2001). Global distribution of perfluorooctane sulfonate in wildlife. Environ Sci Technol 35:1339–1342. https://doi.org/10.1021/es001834k - Hammitt, J.K., Wiener, J.B., Swedlow, B., Kall, D., Zhou, Z. (2005). Precautionary regulation in Europe and the United States: a quantitative comparison. Risk Analysis 25:1215–1228. https://doi.org/10.1111/j.1539-6924.2005.00662.x - Hansen, S.F., Krayer von Krauss, M.P., Tickner, J.A. (2007). Categorizing mistaken false positives in regulation of human and environmental health. Risk Analysis 27(1):255–269. https://doi.org/10.1111/j.1539-6924.2006.00874.x - Horlick-Jones, T., Walls, J., Rowe, G., Pidgeon, N. (2006). On evaluating the GM Nation? public debate about the commercialisation of transgenic crops in Britain. New Genetics and Society 25:265–288. https://doi.org/10.1080/14636770601032858 - Karipidis, K. et al. (2024). The effect of exposure to radiofrequency fields on cancer risk in the general and working population: a systematic review of human observational studies — Part I. Environment International 191:108983. https://doi.org/10.1016/j.envint.2024.108983 - Liu, Y.-Y. et al. (2016). Emergence of plasmid-mediated colistin resistance mechanism MCR-1 in animals and human beings in China. Lancet Infect Dis 16:161–168. https://doi.org/10.1016/S1473-3099(15)00424-7 - Montzka, S.A. et al. (2018). An unexpected and persistent increase in global emissions of ozone-depleting CFC-11. Nature 557:413–417. https://doi.org/10.1038/s41586-018-0106-2 - Owen, R., von Schomberg, R., Macnaghten, P. (2021). An unfinished journey? Reflections on a decade of responsible research and innovation. Journal of Responsible Innovation 8:217–233. https://doi.org/10.1080/23299460.2021.1948789 - Pidgeon, N.F. et al. (2005). Using surveys in public participation processes for risk decision making: the case of the 2003 British GM Nation? public debate. Risk Analysis 25:467–479. https://doi.org/10.1111/j.1539-6924.2005.00603.x - Pilet, J.-B., Bol, D., Vittori, D., Paulis, E. (2022). Public support for deliberative citizens’ assemblies selected through sortition: evidence from 15 countries. Eur J Polit Res 62:873–902. https://doi.org/10.1111/1475-6765.12541 - Qadeer, A. et al. (2024). Global environmental and toxicological impacts of polybrominated diphenyl ethers versus organophosphate esters: a comparative analysis and regrettable substitution dilemma. J Hazard Mater 466:133543. https://doi.org/10.1016/j.jhazmat.2024.133543 - Stirling, A. (2007). Risk, precaution and science: towards a more constructive policy debate. EMBO Reports 8:309–315. https://doi.org/10.1038/sj.embor.7400953 (PMC1852772) - Stirling, A. (2007). A general framework for analysing diversity in science, technology and society. J R Soc Interface 4:707–719. https://doi.org/10.1098/rsif.2007.0213 - Stirling, A. (2010). Keep it complex. Nature 468:1029–1031 (23 December 2010). https://doi.org/10.1038/4681029a (paywalled; content via EPRS 2015 summary) - Tait, J. (2009). Upstream engagement and the governance of science: the shadow of the genetically modified crops experience in Europe. EMBO Reports 10(S1):S18–S22. https://doi.org/10.1038/embor.2009.138 (PMC2725995) - Tang, K.L. et al. (2017). Restricting the use of antibiotics in food-producing animals and its associations with antibiotic resistance in food-producing animals and human beings: a systematic review and meta-analysis. Lancet Planet Health 1(8):e316–e327. https://doi.org/10.1016/S2542-5196(17)30141-9 - Todt, O., Luján, J.L. (2014). Analyzing precautionary regulation: do precaution, science, and innovation go together? Risk Analysis 34(12):2163–2173. https://doi.org/10.1111/risa.12246 - Wang, Y. et al. (2020). Changes in colistin resistance and mcr-1 abundance in Escherichia coli of animal and human origins following the ban of colistin-positive additives in China. Lancet Infect Dis 20:1161–1171. https://doi.org/10.1016/S1473-3099(20)30149-3 - Wesseler, J., Zilberman, D. (2014). The economic power of the Golden Rice opposition. Environment and Development Economics 19:724–742. https://doi.org/10.1017/S1355770X1300065X - Wesselink, A., Paavola, J., Fritsch, O., Renn, O. (2011). Rationales for public participation in environmental policy and governance: practitioners’ perspectives. Environment and Planning A 43:2688–2704. https://doi.org/10.1068/a44161

Books (cited, not re-read) - Graham, J.D., Wiener, J.B. (eds) (1995). Risk vs. Risk: Tradeoffs in Protecting Health and the Environment. Harvard University Press. - Sunstein, C.R. (2005). Laws of Fear: Beyond the Precautionary Principle. Cambridge University Press. - Wiener, J.B., Rogers, M.D., Hammitt, J.K., Sand, P.H. (eds) (2011). The Reality of Precaution: Comparing Risk Regulation in the United States and Europe. RFF Press. - Jordan, G. (2001). Shell, Greenpeace and the Brent Spar. Palgrave. https://doi.org/10.1057/9781403905291 (includes a chapter “The Spar and OSPAR: linking the decisions”)

Secondary (used only where no primary source was retrieved) - Wikipedia, “Brent Spar” (chronology: occupation 30 April 1995; Shell reversal 20 June 1995; Greenpeace apology 5 September 1995; DNV audit 18 October 1995; Mekjarvik reuse announced 29 January 1998). https://en.wikipedia.org/wiki/Brent_Spar