Late Lessons, Jensen Huang and AI
Markdown version

Reception and critique of the EEA Late Lessons reports (2001, 2013)#

Working note, Strand A. Compiled 26 September 2026.

Scope. How the two European Environment Agency (EEA) reports were received and criticised, especially their framing of precaution and “false positives”: what critics argue, which points bear on the reports’ evidence, how defenders replied, how the 2013 report was received, how the “innovation principle” emerged as a counter-frame, and what this means for using the reports as an analytical lens.

Conventions. - LL1 = EEA (2001), Late lessons from early warnings: the precautionary principle 1896–2000, Environmental Issue Report No 22. Page numbers are report pages. - LL2 = EEA (2013), Late lessons from early warnings: science, precaution, innovation, EEA Report No 1/2013. Page numbers are report pages (report page = PDF page − 2). - Both reports were checked against the copies on the EEA website (URLs in the source list).

Access caveats. Faithfulness matters more than coverage, so each source’s access status is flagged: - [full]: the whole text was read. - [abstract]: only the publisher or indexed abstract was read. - [meta]: only bibliographic or publisher-description data was seen. - [known]: the characterisation relies on the work’s well-established thesis and was not re-read for this note.

General web search was unavailable for this task. Several publishers (Taylor & Francis, Wiley, OUP, BMJ, Project MUSE, Duke’s PDF server) returned bot-challenge pages that I did not try to get around. As a result, the 2008 Journal of Risk Research exchange on risk-risk trade-offs and the published reviews of LL1 are known here only from their titles and bibliographic data. Their positions should be checked before they are cited in anything published.

1. Headline findings#

  1. The main critics mostly attack the principle, not the reports. Sunstein, Majone, Marchant, Wiener and Peterson wrote about the principle as such, especially its “strong” versions. Few engage the Late Lessons case evidence in detail. The clearest direct engagements are Marchant (2003) on hindsight and the asymmetry of error verification; Goldstein (2001) on MTBE, itself an LL1 case; Cox (2007) on LL2’s false-positive method; and a hostile Brussels blog (the Risk-Monger, 2013).
  2. LL2’s “4 false positives out of 88 alleged cases” is the most contested single finding. A narrow definition, a demanding threshold for declaring “no harm”, the exclusion of market and de facto effects, and a non-random input list together make genuine false positives hard to find. The authors acknowledged the result is subjective (LL2 p. 33). The EEA’s own messaging then stated it more strongly than the chapter did.
  3. The case-selection and hindsight critique is the strongest structural objection. The reports can show that particular warnings were ignored. They cannot show how often heeding warnings of comparable strength would have been right, because they have no denominator of warnings, including those that turned out false.
  4. Several emerging-issue chapters in LL2 were openly contested. The report printed one industry rebuttal: Bayer’s comments on the neonicotinoids chapter, with the authors’ reply. Later evidence strengthened the bisphenol A (BPA) and neonicotinoid chapters, weakened the mobile-phone chapter, and does not support the report’s concluding claim of health harm from GM crops.
  5. The “innovation principle” was launched in October 2013 by CEOs organised through the European Risk Forum (ERF). It is a procedural counter-frame (“assess the impact on innovation”) that entered Council conclusions in 2016 and the Horizon Europe Regulation in 2021. It reverses LL2’s claim that well-designed precaution stimulates innovation. Evidence on both sides is thin.
  6. Defenders make a coherent reply. Stirling, Gee, Grandjean, Wynne, Tickner and Hansen argue that precaution is a framework for appraisal, not a decision rule; that conventional risk assessment is no less value-laden; that precaution applies to all options, including business as usual; that ignorance differs from uncertainty; and that scientific conventions are biased toward false negatives. This answers the “incoherence” critique well. It is weaker on operationalisation, and on who decides the “appropriate strength of evidence”.

2. What the critics are responding to: the reports’ own claims#

A fair reading starts from what the reports actually say, including their own caveats.

LL1 (2001). - All 14 cases are “false negatives” (LL1 p. 12). The editors disclosed that case authors were mostly “active participants” in the histories they described (p. 12), and asked for judgements based on “the spirit of the times” rather than hindsight (p. 11). - They invited industry representatives to submit false positives. No suitable examples emerged, and the Facts versus fears list was judged not robust enough (pp. 12–13). Candidates mentioned in passing were the North Sea sewage-sludge dumping ban and the “Y2K millennium bug” (p. 13). - LL1 calls the risk of false positives “smaller but commonly feared” (p. 16), argues that a scientific bias toward avoiding Type I errors generates false negatives (p. 184), and claims that curtailing one option can channel innovation elsewhere (p. 182).

Two LL1 features are often missed by critics: - The hormones case contains its own critique of EU precaution. Bridges & Bridges find “no good evidence” the EU ban on hormone growth promoters protected public health, and call the original ban “in reality, a political risk assessment” made against the EU’s own scientific committee. They add that later research “probably justifies” continuing it (LL1 pp. 153–154). - LL1 anticipates the later transatlantic finding. The case histories show “different degrees of acceptance of the need for precaution within different institutions in both North America and Europe”, not an EU–US divide (p. 168).

LL2 (2013). - The introduction says a second volume was needed partly to fill “an acknowledged gap”: the absence of false positives (LL2 p. 9). - It again notes that authors were chosen for their substantial involvement in the cases (pp. 9–10). - It states that false positives “are few and far between as compared to false negatives” (p. 10). - Chapter 2 (Hansen & Tickner) supplies that analysis (pp. 17–45). - Chapter 27 (Gee) offers the EEA’s working definition of the precautionary principle. The definition includes “an appropriate strength of scientific evidence” and “the pros and cons of action and inaction and their distribution” (p. 649). The chapter also gives a scale of evidence strengths for different purposes (Table 27.2, p. 658) and twelve “criteria for action” (Box 27.4, p. 653). - The concluding chapter restates the 4-of-88 result as showing that the risk of false alarms is “considerably less than sometimes claimed” (p. 673).

3. Critiques of the precautionary principle as the reports frame it#

3.1 Incoherence and “paralysis” (Sunstein; Peterson)#

Sunstein [full: Regulation 2002–03; meta: Laws of Fear, 2005]. LL2 cites Sunstein’s short Regulation essay “The Paralyzing Principle” as a representative critic (LL2 p. 18; reference p. 43, where it is misattributed to Reason).

Sunstein sets weak versions aside as unobjectionable and targets strong versions, in which a minimally plausible risk creates a presumption in favour of stringent control. His core argument is that risks lie on all sides. Regulation carries its own dangers: lost “opportunity benefits” (for example delayed drugs), substitute risks (asbestos substitutes, fossil fuels replacing nuclear), and health losses from regulatory costs through the wealth–health relationship. So a strong principle, applied consistently, forbids action and inaction alike.

He explains the principle’s appeal through loss aversion, a “benevolent nature” belief, the availability heuristic, probability neglect and “system neglect”, and notes its strategic use against foreign competitors. He grants some salutary goals: not demanding proof, protecting future generations and the vulnerable. He cites DDT as a case where a ban justified in rich countries may harm poor ones. Laws of Fear develops the argument and proposes a narrow anti-catastrophe principle, cost-benefit analysis and “libertarian paternalism” as alternatives (publisher description; chapter list seen).

Peterson [summary: EMBO Reports 2007] argues that formulations of the principle are too vague to be decision rules, and that a precise version conflicts with basic axioms of rational choice (dominance, Archimedean, total order). He proposes recasting precaution as an epistemic principle about what to believe.

How far this engages the reports. Only indirectly; neither author analyses the cases. The EEA definition, drafted after a decade of this debate, builds in the core of Sunstein’s point by requiring appraisal of “pros and cons of action and inaction and their distribution” (LL2 p. 649), with case-specific strength of evidence (p. 657). Chapter 27 also rebuts critics’ reading of the North Sea Declaration: “no scientific evidence to prove a causal link” does not mean “no evidence” (p. 657).

The critique therefore hits the Wingspread-style “strong” principle much harder than the EEA version. Its residual force: once both sides must be weighed, much of the principle’s work is done by the weighing, which is Sunstein’s point. The EEA definition gives little guidance on how the weighing is done or who does it.

LL2 partly absorbs Sunstein’s behavioural critique. Chapter 25 accepts that public concern can overrate dangers and that rare events, once they occur, may be temporarily overweighted. It cites Sunstein and the post-Fukushima nuclear phase-out decision in Germany (pp. 609, 613).

3.2 Risk-risk trade-offs (Graham & Wiener; Cross; Goldstein; Goklany)#

The 2008 exchange [meta only]. Hansen, Krayer von Krauss & Tickner’s “The precautionary principle and risk-risk tradeoffs” (Journal of Risk Research 11(4): 423–464) drew a comment from Graham & Wiener (465–474), a response from Hansen & Tickner (475–483) and a rejoinder, “Empirical evidence for risk-risk tradeoffs” (485–490). I could not read these texts. Judging from titles and LL2’s summary of the Hansen position (pp. 24–25), the dispute concerns whether trade-offs are pervasive and empirically important (Graham & Wiener) or often overstated and avoidable through alternatives assessment (Hansen & Tickner).

How LL2 treats trade-offs. LL2 files “risk-risk trade-offs” as a category of mistaken false positive, not a failure of precaution. Its nitrite example, where lower nitrite levels plus ascorbate looks like smart risk management rather than over-regulation (pp. 24–25), is defensible. But the classification means the chapter’s headline count cannot register the critics’ main concern: that precautionary action produces new harms. It also sits awkwardly with LL1’s own MTBE chapter.

3.3 Vagueness, arbitrariness and misuse (Marchant; Marchant & Mossman; Majone; Löfstedt; the ERF)#

Marchant [full: EHP 2003]. His main argument is legal. Every jurisdiction that adopted the principle rapidly turned it into a binding rule, and its vagueness then produces arbitrary application by agencies and courts and hampers judicial review. He calls for explicit factors weighing for more or less precaution. Arbitrary and Capricious (Marchant & Mossman, AEI 2004) [meta] makes the case from EU court decisions; LL2 lists it as 2002 (p. 40).

Majone [abstract: JCMS 2002] grants the principle a legitimate but limited role, for example where irreversible damage is imminent. As a general approach, he argues, it lacks a sound logical foundation, can distort regulatory priorities, can be misused for protectionism, undermines international cooperation, may have regressive distributive effects, and, as the Commission interpreted it, applies a double standard between international and intra-EU relations.

LL2’s false-positive table lists “Aflatoxins (Majone, 2002)” and classifies it as a “real risk” (p. 35). As Majone’s aflatoxin example is usually cited [known; not re-checked against his full text], it concerns proportionality and distribution: a very stringent EU aflatoxin standard, a large trade cost to African exporters, and a tiny estimated health gain (cf. Otsuki, Wilson & Sewadeh 2001). Showing that aflatoxins are hazardous does not answer that argument (see section 5.3).

The beef-hormones case is the standard example of alleged protectionism under a precautionary banner. LL1’s own chapter partly concedes the point (pp. 153–154).

Löfstedt [meta]. LL2 lists his (2003) study of the Swedish acrylamide “alarm” as an alleged false positive and classifies it “the jury is still out” (p. 35). Later evidence supports that classification on hazard: EFSA (2015) concluded that dietary acrylamide potentially increases cancer risk, and the EU set mitigation benchmarks (Regulation (EU) 2017/2158). But Löfstedt’s point concerned how the alarm was communicated and handled, which the 88-case framework does not assess. His 2014 paper calls for a formal review of the Commission’s 2000 Communication; I could not read its text.

European Risk Forum [full: Highlights Note 03, December 2015]. Framed as a critique of the principle’s misapplication, the note says it overlooks the costs of false positives, amplifies fear, weakens evidence-based policy, increases discretion and politicisation, is effectively irreversible (because investment stops), institutionalises risk aversion, and in strong forms paralyses risk management and triggers risk-risk effects.

The irreversibility claim bears directly on LL2’s argument that false positives are “short term” because “over-regulation can be quickly caught” (LL2 p. 34). Neither side offers systematic evidence on reversal rates.

3.4 “Selective precaution” and comparative evidence (Wiener, Rogers, Hammitt and colleagues)#

Bearing on the reports. Though not addressed to the EEA, this is the most methodologically pointed critique. It shows that precaution is applied selectively (compatible with LL1 p. 168), that inferences from hand-picked salient cases can mislead, and that which risks get precaution is itself a political outcome needing explanation.

3.5 Fear, trust and the social costs of precaution (Durodié and others)#

Durodié [abstract: Risk Analysis 2003]. Durodié argued that a heavily precautionary chemicals regime (the EU chemicals White Paper) had unrealistic resource implications. He questioned whether a “right to know” always helps, suggesting it can leave people feeling less safe rather than more, and called for social ambition rather than an obsession with safety.

Durodié also reviewed LL1 in Risk Analysis (2002) [meta only]. Other reviews of LL1 appeared in the Journal of Environmental Law (E. Fisher, 2003), Occupational and Environmental Medicine (R. Maynard, 2002) and Public Health (Murray, 2004) [meta only]. Their content should be checked.

4. Case selection and hindsight#

This critique is often voiced but seldom formalised. Its most careful statement, among the sources I could read, is Marchant (2003) [full]. He sets LL1’s list of false negatives (citing Harremoës et al. 2002) against critics’ lists of “excessive precaution”: saccharin, silicone implants, Bendectin, “ice minus”, MMR, swine flu, Bt corn and monarch butterflies, coffee and pancreatic cancer. He then makes four observations:

  1. Trade-off. False positives and false negatives trade off. Better data reduces both, but any setting of precaution buys fewer of one error at the price of more of the other. The real question is whether the balance is optimal, and that is hard to test empirically.
  2. Ignorance, not uncertainty. Many LL exemplars could not have been foreseen at deployment: mesothelioma, ozone depletion, DES adenocarcinoma. The problem there was ignorance rather than uncertainty, and a precautionary principle cannot prevent what is genuinely unanticipated. The stronger charge, which Marchant accepts can be made, is slow response once evidence emerged, or premarket testing too weak to detect harm earlier.
  3. Asymmetry of verification. It is easier to prove harm than to prove its absence. False positives therefore stay provisional and are probably undercounted, while false negatives become settled.
  4. Consequences. False negatives are generally more serious, though not always. Where regulation creates countervailing health risks, the asymmetry may fail. This fourth point is a significant concession to the reports’ position.

Other formal points. Hammitt et al. (2005) add the sampling argument (section 3.4).

Mazur [meta: chapter structure checked via the Internet Archive record]. True Warnings and False Alarms (RFF Press, 2004) is the design that the hindsight critique implies is needed. Mazur took the technology alarms catalogued by Lawless (1977) for 1948–1971, judged them true or false with later knowledge, and coded the “hallmarks” that distinguish true from false alarms. The book closes on “hindsight and foresight”. LL2 used Mazur only as a source of alleged false positives: seven entries in Table 2.3 (pp. 35–36), plus the Southern corn leaf blight case (p. 25). It did not use his prospective-style design. I could not verify Mazur’s quantitative results for this note.

The reports’ own defences against hindsight. LL1 asks for judgement by “the spirit of the times” (p. 11). LL2’s business chapter warns that blaming business “in particular with hindsight” is often not constructive (p. 607; echoed on p. 647). And Chapter 2 excuses two of its own four false positives on ex-ante grounds: it is “not at all obvious”, even with hindsight, that the swine-flu decision was wrong (p. 31), and no one could have known saccharin’s rat-specific mechanism (p. 32). That is the right standard, but it is not applied symmetrically. False positives are excused if they were reasonable ex ante. The false negatives were chosen because harm materialised, and some “early warnings” are dated very early. LL1 begins with an 1898 factory-inspector report on asbestos dust (p. 11).

Assessment. The critique is strong against any frequency or expected-value reading, such as “precaution is nearly always beneficial” or “false alarms are rare”. It is much weaker against the mechanism claims documented case by case: suppression of dissenting science, producer control of research, “manufactured doubt”, externalised costs, regulatory delay, and lock-in of incumbent products. Marchant himself accepts that industry and regulators often responded too slowly once evidence existed.

5. The false-positives debate#

5.1 What Hansen, Krayer von Krauss & Tickner did#

The analysis was published in Risk Analysis (2007a) [abstract] and summarised in LL2 Chapter 2 [full].

Definition (LL2 p. 18). A regulatory false positive is a case where authorities acted on a suspected risk with mandated measures, and there is at least “high confidence” (67–95 % on the IPCC scale) in later evidence that the activity did not pose the suspected risk. Further study, listing a substance, public statements, and market or liability effects do not count (pp. 18–19).

Case list (p. 19). 88 claimed cases, largely from critics’ compilations: Mazur (2004), Wildavsky (1995), Lieberman & Kwon’s Facts versus Fears (1998) and Milloy (2001).

Results (pp. 20–25). About a third were “real risks” and about a third “the jury is still out”. The rest were “unregulated alarms”, “too narrow a definition of risk” or “risk-risk trade-offs”. Four met the definition: Southern corn leaf blight (US, 1971), saccharin labelling (US, 1977), the US swine-flu immunisation (1976), and food irradiation with respect to consumer health.

Conclusions. Fear of false positives “should not be a rationale for avoiding precautionary actions where warranted” (pp. 17, 35). False-positive costs were mainly economic, though swine-flu vaccination caused some deaths (pp. 17, 32). All four cases sparked innovation or learning (pp. 32–33). The scarcity of genuine false positives may partly reflect deliberate industry strategies of “manufacturing doubt” and claiming over-regulation (pp. 17, 33–34).

Stated caveats (p. 33). Categorisation is subjective and other definitions could yield different counts; whether a regulation is excessive is “highly subjective”; some scientists regard phthalate or BPA regulation as over-regulation; readers are pointed to Cox (2007) and the authors’ reply.

5.2 The published critiques#

Cox (2007) [abstract: Risk Analysis 27(5)]. Cox welcomes the empirical idea but argues that the implementation labels many highly uncertain risks “real” on questionable grounds: contamination of a firm’s own product, absence of a known exposure threshold, the mere occurrence of a threat, conservative regulatory assumptions treated as true values, assumed child soil-ingestion treated as proof of dioxin risk, and ambiguous epidemiological associations treated as causal. Such criteria could classify non-existent or unknown risks as real, offering an alternative explanation for the low count.

Hansen, Krayer von Krauss & Tickner (2007b) replied in the same issue [meta only]. LL2 notes the exchange but does not summarise it (p. 33).

Graham & Wiener (2008) [meta only]. Their comment and rejoinder on risk-risk trade-offs (section 3.2) challenge the treatment of trade-off cases as “mistaken” false positives.

The Risk-Monger [full: blog posts of 28 January and 20 March 2013, via the Internet Archive]. This is the openly polemical blog of a Brussels risk-communication commentator (David Zaruk). The March post argues that the five categories were devices for explaining away 84 of the 88 cases; that “unregulated alarms” ignore lasting public fear and market effects; that “the jury is still out” gives dissenters the benefit of the doubt against scientific consensus (naming MMR, GMOs, PVC blood bags, acrylamide and breast implants); that precaution’s real costs come through de facto blacklisting in supply chains, which leaves no regulatory trace (BPA, brominated flame retardants); and that the “manufacturing doubt” passage attacks critics’ motives rather than arguing.

The January post attacks the EEA as an advocacy body, questions some authors’ disciplinary expertise (for example on the bee chapter), and calls the DDT story a precautionary disaster.

5.3 My assessment of the method#

These points are mine, tied to the report’s own text.

  1. The evidential threshold is asymmetric. A precautionary action may be justified by evidence the EEA itself grades “weak (10–33 %)”, such as “scientific suspicion of risk”, or “moderate”, such as “reasonable grounds for concern” (Table 27.2, p. 658). Yet an action counts as a false positive only with “high confidence” (67–95 %) that there is no harm (p. 18). Low bars to act combined with a high bar to count an error guarantee few recorded errors, and tend to leave contested cases in “the jury is still out” indefinitely.
  2. The critics’ main claim was defined out. LL2 describes the critics’ concern as “over-regulation of minor risks and regulation of non-existent risks” (pp. 17–18). The operational definition only captures non-existent risks. Aflatoxins (Majone) are scored a “real risk” (p. 35), which does not answer an argument about proportionality and distribution.
  3. Effects are counted asymmetrically. Market and de facto effects are excluded from the false-positive ledger (p. 18). Yet LL2 treats voluntary market withdrawals as positive signs of precaution, for example user companies abandoning BPA ahead of regulation (p. 647).
  4. There is no denominator. The 88 cases come from polemical lists, many of which were never claims about regulation: American Council on Science and Health items such as “Cheeseburgers and Cardiovascular Disease” or “Soda causes esophageal cancer” (Table 2.3, pp. 35–36). The finding shows that these lists are poor evidence of regulatory over-reach. It does not estimate the false-positive rate of precautionary regulation. LL2’s framing that false positives are “few and far between as compared to false negatives” (p. 10) has no comparable count of false negatives behind it.
  5. Classifications are time-indexed and can move. Acrylamide moved toward “real” (EFSA 2015). For mobile phones, the chapter’s main “jury is still out” example (pp. 21–22), later large studies point the other way (Schüz et al. 2022; the WHO-commissioned review by Karipidis et al. 2024). Yet it could never become a regulatory false positive, because no binding regulation was adopted. By construction, the framework cannot register a large class of false alarms: those that generated fear, research and market effects but no regulation.
  6. The finding was amplified. - The EEA’s launch release said a precautionary approach is “nearly always beneficial”, citing only four clear cases among 88 (EEA press release, 22 January 2013). - LL2’s conclusion restates it without the chapter’s caveats (p. 673). - The European Parliament’s research service reproduced the 4-of-88 result as showing little risk of false positives (Bourguignon 2015). - The Guardian’s launch story garbled it, reporting that the authors had found over 80 cases where regulation was not introduced although the risk later proved real or remained unproven (Vidal 2013).

Net judgement. Hansen and colleagues made a genuine and useful contribution. They showed that the standard lists of “scares” used to argue against precaution are unreliable, and they set out transparent categories others can contest. But the result does not support a general claim that precautionary regulation seldom errs. The fairest reading is narrower: the strongest claimed examples of over-regulation mostly turn out, on inspection, to be real risks, unresolved risks, or never regulated.

6. Reception of the 2013 report#

The EEA’s framing at launch [full: press release]. The EEA presented the report as a 750-page volume with 20 new case studies. It stressed that warnings were suppressed or ignored, that “in some instances, companies put short-term profits ahead of public safety”, and that precaution can “stimulate rather than stifle innovation”. It also claimed that faster, global technology uptake means risks may now outstrip society’s capacity to respond.

Media [full: Guardian, 23 January 2013]. John Vidal’s report led with Europe’s failure to learn from past disasters. It foregrounded the chapters most likely to be contested: the Chernobyl death estimates (“at least 17,000 to 68,000 over 50 years”, against the WHO’s much lower figures), a claimed 20-fold increase in the probability of a major nuclear accident, the GM chapter’s view that benefits are overstated, and the “manufacturing doubt” passage. The UK Science Media Centre site has no expert-reaction page on the report.

Industry responses printed in, or alongside, the report. - Neonicotinoids (Ch. 16). This is the most formal rebuttal. Bayer CropScience’s Richard Schmuck contributed Panel 16.1 (pp. 401–402), with a longer statement in a separate EEA “bee decline debate” document. Bayer argued that the authors used selective data, discounted the multi-factor French AFSSA assessment (2009), relied on a superseded French risk-assessment scheme, misrepresented the EU system, and ignored monitoring that found no correlation between colony losses and seed-dressing exposure.

Maxim and van der Sluijs replied (Panel 16.2, pp. 403–404) that the French decisions were slow, not “rapid”; that multicausality is no reason to ignore a controllable cause; that seed-drilling dust incidents continued and fell sharply under the Italian moratorium; and that Bayer’s pre-registration reasoning had extrapolated from aphids to bees.

Later evidence moved toward the chapter. EFSA’s 2018 peer reviews confirmed risks to bees, and the EU restricted imidacloprid, clothianidin and thiamethoxam to permanent greenhouses (Implementing Regulations (EU) 2018/783–785). This undercuts the Risk-Monger’s 2013 claim that colony collapse had simply been shown to be a viral problem. - Mobile phones (Ch. 21). The chapter quotes the Mobile Manufacturers Forum and the GSM Association responses to IARC’s 2011 “possibly carcinogenic” classification (p. 520). These were responses to IARC, not to the EEA.

Three features matter for weighing it. It is co-authored by Lennart Hardell, whose group’s studies are half the evidence base it assesses (p. 509), and by the report’s editor, David Gee. A footnote records that the EEA withdrew from the IARC meeting, citing delays to the Interphone results and the “intellectual bias” of the then chair (p. 520), which places the agency inside the controversy. And later evidence (Schüz et al. 2022; Karipidis et al. 2024) does not support the association the chapter treated as a probable early warning. - GM crops (Ch. 19; Ch. 28, p. 674). The concluding chapter states that there is evidence some GM crops “present a threat to human health”. The US National Academies’ 2016 review found no substantiated evidence that approved GE foods are less safe than conventional ones. It did report agronomic and environmental problems such as herbicide-resistant weeds. The health claim is the weakest-supported synthesis statement in LL2. The chapter’s broader argument, about innovation pathways and research-funding imbalance, is a different and more defensible claim. - BPA (Ch. 10). LL2 itself acknowledges that some scientists regarded BPA restrictions as over-regulation (p. 33). Later evidence moved decisively toward the chapter. EFSA (2023) cut the tolerable daily intake 20,000-fold, and the EU banned BPA in food-contact materials (Regulation (EU) 2024/3190). - DDT (Ch. 11). DDT and malaria is a staple of critics (Sunstein 2002–03; Goklany 2001; the Risk-Monger 2013). LL2 classifies “DDT and malaria” as a “real risk” in Table 2.3 (p. 36). The critics’ argument is distributive: a ban suited to rich countries may cost lives in poor ones. That argument is not about whether DDT is hazardous, so the classification does not meet it. - Nanotechnology (Ch. 22) is co-authored by Andrew Maynard, with Hansen, Baun, Tickner and Bowman. The chapter states that it is based on, and partly identical to, a 2008 Nature Nanotechnology commentary (p. 530). I found no formal published rebuttal of it. It has been built on in later nanomaterials safety-by-design work (for example Brennan & Valsami-Jones 2021). In 2016 the ERF argued against “horizontal” nano-specific rules in general, but not against this chapter.

Critics’ commentary. The Risk-Monger’s two 2013 posts are the only sustained hostile commentary I found (section 5.2). They matter less for their arguments, several of which are ad hominem or have been overtaken by later evidence, than as a record of how the report was read in Brussels policy circles: as advocacy by an EU agency.

EU institutions. I found no formal Commission response to LL2. The European Parliament’s research service (Bourguignon 2015) [full] presented the debate evenly, setting critics (Marchant & Mossman, Majone, Sunstein) against those who see the principle as useful for complex hazards; it drew heavily on LL2 (Figure 2, p. 14; pp. 16–17) and reported the ERF’s innovation-principle proposal. DG Environment’s Future Brief (Science for Environment Policy 2017) [full] repeats standard criticisms (ill-definition, Sunstein’s “paralysis”, anti-science, anti-innovation), pairs them with responses from Stirling and Fisher, and adds a box on the innovation principle.

Scholarly uptake by the reports’ authors. - Hansen & Gee (2014, JECH) [abstract] extended LL2’s analysis of research imbalance. They found that environmental, health and safety research was about 0.6 % of EU research and technological development funding since 1996. - Hansen and colleagues (2024, Nature Water) applied the approach to PFAS.

7. The “innovation principle”#

Origin [full: ERF letter of 4 November 2014; Monograph; Highlights 03 and 07]. - On 9 October 2013, CEOs coordinated by the European Risk Forum wrote to Presidents Barroso, Van Rompuy and Schulz proposing an “Innovation Principle”. - A second letter to President Juncker, dated 4 November 2014, defines it: whenever policy or regulatory decisions are under consideration, their impact on innovation should be fully assessed and addressed. - The letter says the signatories’ firms invest €30 billion in innovation. It objects to a regulatory climate that, in its view, increasingly avoids risk instead of managing it. It asks for the principle to be placed in the Commission’s Better Regulation and Impact Assessment work. - Signatories included BASF, Bayer, Dow, DuPont, Evonik, Novartis, Solvay, Syngenta and others. Bayer and Syngenta were among the neonicotinoid manufacturers. - The ERF monograph links slower EU productivity growth to a regulatory culture it describes as favouring social concern over science. - A 2016 ERF note gives the 2013 neonicotinoid restrictions as an example of poorly designed risk management. - The ERF’s 2015 note on precaution frames its critique as concerning the principle’s misapplication.

Uptake. - Council. The Competitiveness Council conclusions of 27 May 2016 called for the principle to be applied when developing or updating EU measures (as reported in Science for Environment Policy 2017, Box 2). - European Political Strategy Centre. The Commission’s in-house think tank published a 2016 Strategic Note (No 14). As summarised in the same Future Brief, it stressed that the precautionary principle is vital for innovation: it supplies procedures to assess and control risks while possible risks cannot be eliminated. - Legislation. Recital 6 of the Horizon Europe Regulation (Regulation (EU) 2021/695) asks funded activities to take advantage of innovation-friendly regulation “in line with the innovation principle”. The regulation text I checked does not mention the precautionary principle.

Critiques. - Corporate Europe Observatory (NGO, 2018) [full]. The group presents the principle as an industry invention with no Treaty basis, promoted by a lobby group drawn from heavily regulated sectors (chemicals, pesticides, fossil fuels, and originally tobacco). It documents the principle’s path from a 2015 Commission mention, through the 2016 Council conclusions, to Horizon Europe. It quotes the Leuven legal scholar Geert van Calster’s concern that repeated insertion into documents is meant to make it seem an established principle. - Garnett, Van Calster & Reins (2018, Law, Innovation and Technology) [meta]. The title asks whether the principle is an “industry trump” or could favour environmental protection. - Stirling (2016) [full]. Stirling treats such principles as legitimate lobbying interventions but loosely specified, and not comparable with forms of precaution built up over decades of negotiation and judicial practice. His substantive point is that precaution concerns the reasons for intervening, not their stringency, and that curbing one technological trajectory advantages others. Precaution, in his phrase, is about “steering, not stopping” innovation.

Relation to Late Lessons. The innovation principle mirrors LL2’s innovation thesis: that precautionary actions, even unnecessary ones, can spark innovation (Lesson 6, p. 35; pp. 32–33); that late action entrenches incumbent products and keeps substitutes out (p. 659); and that “potential for stimulating innovation” is a criterion for action (Box 27.4, p. 653; cf. LL1 p. 182). LL2’s evidence is mostly illustrative (saccharin and alternative sweeteners, the swine-flu surveillance legacy, Porter and Ashford on regulation-induced innovation). The ERF’s is at a similar level (macro productivity comparisons, selected cases). Neither side offers counterfactual evidence on how much innovation a given precautionary act prevented or induced. The dispute is really about which procedural default should frame impact assessment, which is why both principles now coexist in EU documents.

8. Defenders and developers of the reports’ approach#

Andy Stirling [full: SPRU 2016; summary: EMBO Reports 2007], an LL1 editor, argues that precaution is a normative framework for broadening appraisal (independence, blind spots, alternatives, participation), not a decision rule. Applying risk methods to uncertainty, ambiguity and ignorance is itself the “unscientific” move, and because risk assessment and cost-benefit analysis also embed values, the principle’s open normativity is more accountable, not less. He concedes that precaution can be misapplied selectively, for example imposed on some options and not on business as usual, but calls that a fault of application. Critics and proponents, he says, share the goal of more rigorous, symmetric appraisal, a concession that overlaps substantially with Sunstein’s and Marchant’s demands.

Brian Wynne [meta] was also an LL1 editor. He supplied the conceptual basis for LL’s distinction between uncertainty, ignorance and indeterminacy (Wynne 1992). With Hoffmann-Riem (2002) he argued that risk assessment must admit ignorance. Marchant cites that argument to make the opposite point: the principle cannot prevent the unanticipated. The two readings are compatible. Ignorance argues for monitoring, diversity and reversibility (LL1 lessons 1, 2 and 7, pp. 168–169) more than for ex-ante prohibition.

David Gee [abstracts: EHP 2006; Pathophysiology 2009; full: LL2 Ch. 27] develops differential strengths of evidence for different purposes, drawing on Bradford Hill’s call for action (LL2 pp. 656–658); the asymmetric use of the Bradford Hill criteria under multicausality, where absent criteria are not evidence of no effect (pp. 653–654); and methodological and funding biases toward false negatives (p. 654). His chapter frames resistance to the principle largely as vested interests plus conservative science (p. 643). Critics read that framing as dismissive of legitimate disagreement.

Philippe Grandjean [abstract: Epidemiology 2008; full: LL2 Ch. 26 summary] argues that conventions such as treating statistically non-significant results as “negative” bias research toward false negatives. He also argues that research concentrates on well-studied agents (LL2 p. 623).

Joel Tickner and Steffen Foss Hansen developed the false-positive analysis and the risk-risk rebuttal (sections 5 and 3.2). They advocate alternatives assessment as the practical answer to trade-offs (LL2 p. 35, Lesson 3).

Philosophers. Sandin, Peterson, Hansson, Rudén & Juthe (2002) [meta/known] argued that the standard charges (ill-defined, absolutist, risk-increasing, ideological, unscientific) can be answered. Daniel Steel (2015) [known] builds a version designed to meet the incoherence objection through proportionality and consistency requirements. These are the most direct answers to Sunstein and Peterson, but they defend a principle, not the EEA’s case evidence.

9. Assessment: strong and weak critiques, and implications for using the reports as a lens#

9.1 Which critiques are strong#

  1. Case selection and missing base rates (Marchant; Hammitt et al.; Mazur’s design). Strong against any frequency or expected-value reading. The reports’ own disclosures confirm the selection: authors chosen for involvement (LL1 p. 12; LL2 pp. 9–10) and cases chosen because harm is now known (LL1 p. 11).
  2. The false-positive method (Cox; the definitional and threshold points in section 5.3). Strong against the headline “4 of 88” as a general rate. It remains a fair and useful rebuttal of critics’ stock lists.
  3. Risk-risk trade-offs and distribution (Graham & Wiener; Goldstein; Majone; Sunstein on DDT). Moderately strong. LL documents such trade-offs itself (MTBE, LL1 ch. 11) and endorses alternatives assessment. But its categorisation keeps them out of the ledger of precautionary error.
  4. Advocacy and positioning in the contested chapters. Moderately strong for Ch. 21 on mobile phones, given the author’s involvement and the EEA’s IARC footnote. It also applies to the GM health claim in Ch. 28, and to the nuclear figures that received the most media attention. These are exactly the chapters where later evidence has been least kind.
  5. Legal vagueness (Marchant; Marchant & Mossman). Strong in practice. The EEA definition’s key terms, “appropriate strength of evidence” and “pros and cons … and their distribution”, are procedural placeholders. They move the hard choices elsewhere without resolving them.

9.2 Which critiques are weak or only partly relevant#

  1. “Paralysis” and incoherence (Sunstein; Peterson). This is decisive against strong versions, but largely met by the EEA’s two-sided, case-specific definition (LL2 p. 649). Its remaining force concerns operational thinness, not logic.
  2. “Anti-science” and “anti-innovation” as blanket charges. Weak. LL1 and LL2 argue for more and different science: monitoring, attention to multicausality, and funding for hazard research. They argue for redirecting innovation, not against it. The ERF itself frames its critique as one of misapplication.
  3. Ad hominem and expertise objections (Risk-Monger). Weak as argument. The bee chapter is a governance analysis, and on its substantive question later evidence moved toward it. These objections are still useful as evidence of how the reports were received.
  4. “Europe is more precautionary” as a framing used by both sides. This is undermined by Hammitt et al. and The Reality of Precaution. It is a point LL1 had already conceded (p. 168).

9.3 Where the critics themselves concede ground#

Marchant accepts that false negatives are generally more serious and that responses were often too slow once evidence existed. Majone accepts a role for precaution where irreversible damage is imminent. Sunstein accepts the value of not demanding proof and of protecting the vulnerable. On the other side, Stirling accepts that selective or opaque application is illegitimate.

9.4 Implications for using the reports as an analytical lens#

These are stated in technology-neutral terms, each tied to the reports.

  1. Use the reports for mechanisms, not frequencies. Their robust contribution is a catalogue of recurring dynamics documented across several cases: producer control of research and “manufactured doubt” (LL2 Ch. 7; p. 646); costs externalised onto third parties and future generations (p. 647; Ch. 23); institutional inertia and regulatory delay (pp. 34, 645); “no evidence of harm” confused with “evidence of no harm” (p. 658); research skewed toward known hazards and product development (pp. 623, 646); and lock-in of incumbent products that keeps substitutes out (p. 659). None depends on the false-positive rate. Avoid relying on “false alarms are rare” or “precaution is nearly always beneficial” (p. 673; the 2013 press release).
  2. Always pair each lesson with the critics’ countervailing mechanisms. Using the lens properly means asking of any proposed precautionary response whether it: - creates substitute or countervailing risks (LL1 ch. 11; LL2 pp. 24–25); - forgoes benefits, especially for the least advantaged (Majone; Sunstein on DDT); - can be captured to protect incumbents or markets (Majone; LL1 pp. 153–154); - is reversible in practice (ERF 2015; LL2 p. 34).
  3. Judge ex ante, and do it symmetrically. Ask what was knowable, by whom, and at what strength of evidence at the time, and what the warning was worth given the base rate of comparable warnings. LL2’s own excusal of the swine-flu and saccharin decisions (pp. 31–32) models the right standard. The lens should apply it to both error types, and should apply LL1’s lesson 6 (scrutinise claimed benefits alongside risks, p. 169) to critics’ benefit claims as well as to producers’.
  4. Distinguish by evidential grade within the reports. - The historical false-negative cases (for example lead, asbestos, tobacco, vinyl chloride, DBCP, beryllium) are well evidenced. - Some emerging-issue chapters have been strengthened by later evidence (BPA, neonicotinoids). - Others are contested or weakened (mobile phones, the GM health claim, nuclear casualty figures). - The synthesis chapters (Ch. 27 and Ch. 28) are explicitly advocacy for wider use of the principle and should be read as such.
  5. Treat selective precaution as the phenomenon to explain. The comparative literature (Wiener et al.) and LL1 (p. 168) agree that precaution is applied unevenly by institution and by risk. For a lens, the useful question is less “precaution: yes or no?” than who triggers precaution, for which harms, at what evidence threshold, and who bears the costs of acting and of not acting (the EEA definition’s distributional clause, p. 649).

Sources#

The reports and associated EEA material#

Critics and comparative work#

The false-positives debate#

Reception, 2013 onward#

Innovation principle#

Defenders and developers#

Later evidence cited for contested chapters#