Late Lessons, Jensen Huang and AI
Markdown version

LL2-24 hindsight check: Ch24 “Protecting early warners and late victims”#

Late lessons from early warnings: science, precaution, innovation (EEA Report No 1/2013), report pp. 581–606 (PDF pp. 583–608). Author: Carl Cranor. Panels by David Gee (24.1), Owen McIntyre (24.2) and a summary of Robert Costanza (24.3).

Check date: 26 September 2026.

Method note. The general web-search budget for this session ran out before this section was started. Every source below was therefore retrieved directly from primary repositories: EUR-Lex/Publications Office (Cellar), Légifrance, legislation.gov.uk, UK Find Case Law, First Circuit opinions, the S.D. Ill. MDL docket page, eCFR, govinfo, SEC EDGAR, EPA, IARC, PubMed, and the websites of the schemes and prizes concerned. Some points that would normally be checked by search could not be verified. Each of these is marked [not verified in this pass], and none carries weight in a verdict.

This chapter is new in the 2013 report, so there is no Annex 3 update for it.

Overview#

Chapter 24 is a normative legal-policy essay, not a case study. Its two proposals are: - stronger protection for people who give early warnings; - faster compensation for victims that does not depend on proving fault.

Thirteen years on, the record splits cleanly between the chapter’s structural diagnosis and its specific proposals.

The structural diagnosis has been borne out repeatedly. The chapter argues four things: - proof rules decide who carries scientific uncertainty; - tort is slow and expensive for claimants; - tables of presumed injuries need a history of earlier victims; - latency and firms that disappear defeat compensation after the fact.

Later events fit all four: - Milward ran from 2007 to 2016 and ended with the plaintiffs losing. - The paraquat litigation (over 5,000 plaintiffs) went from exclusion of the plaintiffs’ causation expert in 2024 to a tiered “settlement matrix” in 2025. - Camp Lejeune compensation for contamination in 1953–87 was legislated only in 2022, under an “at least as likely as not” standard, and payouts are still arriving in 2026. - The UK created a levy-funded scheme in 2014 specifically for mesothelioma victims who cannot trace a solvent employer or insurer.

The specific proposals were mostly not adopted in the form proposed: - The EU Whistleblowing Directive (2019/1937) contains several features Cranor asked for, including reasonable belief, secure channels and a reversed burden of proof on retaliation. But it protects reports of breaches of Union law, not scientific warnings about lawful but hazardous activity. - France’s 2013 health and environment alert commission, the closest match to the chapter’s recommendation, was abolished in 2026. - No “Henrik Ibsen award” appeared. The German whistleblower prize the chapter cites went in 2015 to a scientist whose GMO rat study had been retracted, which illustrates the “false alarm” cost the panel itself acknowledged. - No general no-fault scheme was created for chemicals, GM crops, nanomaterials or mobile phones. Presumptive schemes did grow in narrower forms: - occupational pesticide-related Parkinson’s disease (France 2012 and 2020; Germany BK 1322); - pollution caused by government (Camp Lejeune); - “orphan” asbestos victims (UK 2014). - Worst-case assurance bonds were not adopted anywhere I could find. The nearest real change is the EU 2013 Offshore Safety Directive’s requirement that licence applicants make provision to cover liabilities.

The empirical examples have fared unevenly:

Example How it fared
Trichloroethylene (TCE) and Parkinson’s disease Strengthened: large 2023 cohort study; US presumptive compensation; 2024 US ban
Paraquat and Parkinson’s disease Remains contested
Radiofrequency (mobile phone) exposure and brain tumours Weakened substantially (2024 WHO-commissioned review)
Bisphenol A: proving it causes disease is “extremely difficult” Held up. The EU acted in 2024 on a hazard basis rather than on proven human disease

Two errors and one undisclosed interest matter for how much weight the chapter can bear: - The Deepwater Horizon liability cap was USD 75 million plus all removal costs, not USD 50 million. - The Vaccine Injury Compensation Program (VICP) arithmetic is wrong by a factor of 10. - Cranor was the plaintiffs’ expert on scientific methodology in Milward, the case the chapter holds up as marking a change. The chapter does not say so.

Claim-by-claim#

1. Extend whistleblower protection to early-warning scientists (p. 584)#

Original claim (p. 584; see also pp. 582–583). Whistleblower laws should be extended to protect scientists and others, in the public and private sectors, who warn of threats to health or the environment. The recommended design: - protection that rests on a reasonable belief, not on being proved right; - a secure reporting channel; - an easy prima facie case of retaliation; - an independent office to protect and prosecute; - optional bounties paid as a share of fines.

The models are the US Whistleblower Protection Act (WPA) and SEC rules, the UK Public Interest Disclosure Act (PIDA) and South African law, all taken from one 2001 source (Kaplan).

Subsequent developments

Its material scope is limited to “breaches of Union law” in listed areas, including product safety, environment, radiation protection, food and feed safety, public health and consumer protection (Art. 2). The Annex covers REACH and the pesticide and food-law acts. A “breach” is an act or omission that is unlawful or that “defeat[s] the object or the purpose of the rules” (Art. 5(1)).

The Directive therefore protects a scientist who reports suppressed test data or non-compliance. It does not, by itself, protect a scientist who warns that a lawfully authorised product is hazardous. That is the core “early warning” situation in Late Lessons. Member states may extend the scope (Art. 2(2)).

The Directive contains no bounties. It requires designated “competent authorities” but not an independent office to protect and prosecute of the kind Cranor describes. - Implementation was slow and uneven. The Commission’s report of 3 July 2024 found [S2]: - only 3 member states had transposed on time (deadline 17 December 2021); - infringement proceedings were opened against 24 states in January 2022; - “several Member States” had omitted “reasonable suspicions” and breaches “very likely to occur”; - “several” had mis-transposed the reversed burden of proof, “by requiring the reporting person … to establish prima facie that a certain measure was taken in retaliation”; - “A large majority” extended the scope nationally, and “several” extended it to “all (serious) breaches of national law that may harm the public interest”.

The Court of Justice fined three member states:

Case Member state Date Penalty
C-147/23 Poland 25 April 2024 EUR 7 million lump sum plus EUR 40,000 a day [S3]
C-149/23 Germany 6 March 2025 EUR 34 million [S4]
C-150/23 Luxembourg 6 March 2025 EUR 375,000 [S5]

Article 27(3) required an impact evaluation by 17 December 2025, and the Commission said “no later than 2026”. On 26 September 2026 the Commission’s whistleblower page showed no such evaluation [S6]. - France came closest to Cranor’s proposal, then reversed. - Loi n° 2013-316 of 16 April 2013 (the “loi Blandin”), passed a month before this report, created a right to make public information about serious risks to public health or the environment. It also created a Commission nationale de la déontologie et des alertes en matière de santé publique et d’environnement (cnDAspe) [S7]. - The Sapin II law (2016, amended 2022) generalised protection to anyone reporting “une menace ou un préjudice pour l’intérêt général” (a threat or harm to the general interest). This is broad enough to cover scientific warnings. It requires the report to be made “sans contrepartie financière directe”, which rules out bounties [S8]. - Loi n° 2026-403 of 26 May 2026 (“de simplification de la vie économique”, Art. 1(V)) repealed Article 2 of the 2013 law, abolishing the cnDAspe, effective two months after promulgation. What remains is a duty on public expert bodies in health and environment to keep registers of alerts, open to the Sapin II authorities [S7][S9]. - UK. The PIDA categories of disclosure already included danger to “health or safety” and to “the environment” (Employment Rights Act 1996, s.43B(1)(d)–(e)) [S10]. From 25 June 2013 the Enterprise and Regulatory Reform Act 2013 s.17 added the condition that the worker must reasonably believe the disclosure is “in the public interest” [S11]. The UK still has no independent whistleblower office [not verified in this pass]. - An independent office. The Dutch Whistleblowers Authority, created under a 2016 Act and continued under the 2023 Act, investigates both wrongdoing and “reprisals against reporting persons”. This is close to Cranor’s model [S12]. - US. - The Whistleblower Protection Enhancement Act (signed 27 November 2012, too late for the chapter) made it a protected disclosure to report “censorship related to research, analysis, or technical information”, defined as “any effort to distort, misrepresent, or suppress research, analysis, or technical information”, where it threatens public health [S13]. - SEC bounties continue in the financial field: over USD 60 million to 48 whistleblowers in FY2025 [S14]. I found no bounty scheme aimed at scientific warnings about health or environmental hazards. - In 2025, staff at the EPA, NIH and NASA published formal dissent declarations (secondary source [S15]). Their consequences were not verified here. This at least questions the chapter’s assumption that retaliation against government scientists is “perhaps … less likely” (p. 582).

Verdict: partly held up. Whistleblower protection expanded greatly, and the EU adopted several of Cranor’s design features. But the central extension he asked for, to cover scientific warnings about lawful activity, is at best a national option; the EU standard excludes it. The one bespoke body built for health and environment alerts (France’s cnDAspe) lasted from 2013 to 2026. Bounties stayed confined to financial enforcement.

Implications for weight. The diagnosis holds: law protects employees who report illegality much better than scientists who report hazard. The Directive confirms, rather than closes, the gap Gee identifies in Panel 24.1. That the recommendation was not taken up shows that the institutional appetite for protecting “early warners” as a class is limited and reversible, not that the idea was wrong. Treat the design list as a sound specification of a protection regime. Do not treat it as a description of what exists.

2. Professional associations should defend harassed early-warning scientists, and a European “Henrik Ibsen award” should be created (Panel 24.1, p. 585)#

Original claim (Gee, pp. 584–585). - Independent professional associations should adopt “explicit policies that encourage early warning scientists and defend them if they are harassed”. - An authoritative, unbiased scientific body should award a European “Henrik Ibsen award for early warning scientists”, on the model of the German Whistleblower Prize (VDW/IALANA). - Gee accepts that some warnings “turn out to be false alarms” and calls this “an acceptable price”.

Subsequent developments - Ibsen award. No award by that name, and no EU-level award for harassed early-warning scientists, was found [not verified in this pass: the absence could not be confirmed by open search]. - The German Whistleblower Prize is still described on the VDW site (page updated 4 March 2026). Its list of laureates “since 1999” begins with 2017, the most recent year listed [S16]. The 2015 laureates included Gilles-Éric Séralini [S16]. His 2012 study of a Roundup-tolerant GM maize, published in Food and Chemical Toxicology, is indexed in PubMed as a “Retracted Publication” [S17]. The prize the chapter recommends as a model thus honoured a contested warning. This is exactly the false-alarm risk Gee accepted. - The John Maddox Prize is the nearest analogue that exists. It has been run since 2012 by Sense about Science, with Nature, and recognises people who “stand up for science and evidence … despite challenges or hostility”. It has an early-career category and states that “we don’t yet live in a world where it is safe for researchers to speak out” [S18]. Its orientation differs from the proposed award: it rewards defending sound evidence under hostility, not challenges to scientific orthodoxy as such. - Codes of conduct. The ALLEA European Code of Conduct for Research Integrity, revised in 2023, now says: “Research institutions and organisations actively support researchers who receive threats and protect bona fide whistle-blowers, taking into account that early career and short-term employed researchers may be particularly vulnerable” [S19]. This is institution-level, framed around integrity rather than “early warnings”. It is still the closest European uptake of the panel’s idea. - Support services. Legal and practical support services now exist: - the US Climate Science Legal Defense Fund, founded 2011, which provides “free legal advice … to academic and government researchers targeted” [S20]; - Germany’s Scicomm-Support, a national contact point for attacks on researchers [S21]; - the Netherlands’ WetenschapVeilig, a central reporting point for threatened scientists at Dutch universities and KNAW/NWO institutes [S22].

Founding dates for the last two were not verified.

Verdict: partly held up. The problem Gee described gained recognition, and support infrastructure now exists, mostly in the form of integrity codes, support services and a prize for defending evidence. His specific proposals did not materialise: policies that explicitly defend “early-warning” dissenters, and an authoritative European award. The one existing model he pointed to shows the selection risk in practice.

Implications for weight. - The insight that people whose own work is challenged are conflicted judges of the challenge (Gee’s use of the WTO hormones ruling) is untouched. - The recommendation’s feasibility is weak. Honouring dissent before it is vindicated means honouring some dissent that is not. - Later analysis should treat “recognition of warners lags vindication” (p. 584) as sound, and “award the warners” as a contested remedy.

3. Post-Daubert rules “asymmetrically hamper plaintiffs”; Milward v Acuity (1st Cir. 2011) marks a change toward weight-of-evidence testimony (p. 588)#

Original claim (p. 588). - Daubert, Joiner and Kumho “in principle … impartially apply to both plaintiffs and defendants, [but] in reality … asymmetrically hamper plaintiffs”. - Milward “marked a change”: scientists in the First Circuit may now use in court “the same kinds of arguments that they would use in the lab”, and there is “no priority of evidence” such as epidemiology for cancer causation. - In 2012 the case was ongoing against one defendant.

Subsequent developments - Undisclosed interest. The 2011 First Circuit opinion records that the evidentiary hearing heard “plaintiffs’ experts Dr. Smith, a toxicologist, and Dr. Carl Cranor, an expert on scientific methodology”. The court adopted Cranor’s six-step account of “inference to the best explanation” [S23]. The chapter praises the ruling without disclosing this role. - What happened to Milward after remand. A different district judge excluded the plaintiffs’ specific-causation expert and granted summary judgment to the last defendant, Rust-Oleum. On 25 April 2016 the First Circuit affirmed (2–1, Thompson J. dissenting). It found no abuse of discretion given the “extraordinary number of idiopathic APL cases, coupled with the lack of a reliable means to rule out an idiopathic diagnosis” (APL is acute promyelocytic leukaemia) [S24]. The case filed in 2007 ended nine years later with no recovery from the remaining defendant. This is strong confirmation of the chapter’s separate point that tort is slow and uncertain (p. 588). - Federal Rule of Evidence 702 was amended, effective 1 December 2023. Admissibility now requires the proponent to demonstrate “it is more likely than not” that each reliability requirement is met. The Advisory Committee note states that holdings treating “the sufficiency of an expert’s basis, and the application of the expert’s methodology” as “questions of weight and not admissibility … are an incorrect application of Rules 702 and 104(a)”. The Committee also says the amendment “imposes [no] new, specific procedures” [S25]. In practice this moves in the opposite direction to the liberalisation Cranor welcomed. - Weight of evidence was accepted in principle and tightened in application. In the paraquat multidistrict litigation (MDL) (S.D. Ill., order of 17 April 2024) [S26]: - the court said “the general reliability of this approach is not in dispute”, citing Milward and In re Zoloft (3d Cir. 2017); - it also held, following Zoloft, that an expert must explain “how the criteria are weighed relative to one another”; - it excluded the plaintiffs’ sole general-causation expert, calling his analysis “a textbook example of the type of standardless presentation of evidence that courts have cautioned against”; - it applied the amended Rule 702 and cited In re Acetaminophen (S.D.N.Y. 2023), which applied the amendment to exclude plaintiffs’ experts.

Milward’s “reasoning to the best explanation” survives as an accepted method. The practical bar is now a transparent, weighted application, and the amendment reinforced the judge’s gatekeeping role. - The other side of the argument. In the paraquat order the court relied partly on EPA’s 2019 systematic review (“limited, but insufficient epidemiologic evidence”) and on a 2021 “review of reviews” asserting a scientific “consensus” against causation (see claim 5). On this view, exclusion reflects weak evidence rather than bias against plaintiffs. Whether the rules “asymmetrically hamper” plaintiffs therefore remains a question of values and law as much as fact.

Verdict: partly held up. Burdens on plaintiffs’ experts did not ease after 2013, and on the federal side they arguably increased. That is consistent with the “asymmetric” effect the chapter describes. The prediction that Milward “marked a change” was only half right: weight-of-evidence reasoning spread as a recognised method but was curtailed in application. Milward itself ended in defeat.

Implications for weight. - The mechanism that formally neutral gatekeeping disadvantages whoever carries the burden of proof remains well supported as a structural observation. - The chapter’s optimistic reading of Milward should be discounted. The author’s role in the case should be noted whenever the chapter’s legal claims are relied on.

4. UK asbestos law (Fairchild, Margereson) “could illustrate the future direction of long-tail liability” and might extend to environmental damage (pp. 581, 590–593)#

Original claim (Panel 24.2, pp. 591–593; summary p. 581). - Fairchild (2002) relaxed “but for” causation for mesothelioma and, the panel says, created joint and several liability. - Margereson (1996) made liability to neighbours of an asbestos factory depend on foreseeing “some pulmonary injury”, not mesothelioma specifically. - The panel suggests these rules “could illustrate the future direction of long-tail liability in both environmental damage and personal injury”, with foreseeability judged by “broad classes of damage”.

The panel is essentially from 2004 and omits the 2006 developments.

Subsequent developments - Apportionment was overruled by Parliament. Barker v Corus [2006] UKHL 20 held that Fairchild liability is proportionate to each defendant’s share of risk. Within weeks Parliament reversed this for mesothelioma. Compensation Act 2006 s.3 makes a responsible person liable “in respect of the whole of the damage … jointly and severally” [S27]. In Sienkiewicz v Greif [2011] UKSC 10 the Supreme Court restated the rule, noting that “the rejoicing with which the insurance industry must have greeted” Barker “was short lived as Parliament intervened” [S28]. - Scope. Sienkiewicz applied Fairchild where a single employer’s exposure added only 18% to the victim’s environmental background exposure. The Court refused to require proof that the exposure “doubled the risk”. Lord Brown concurred but warned of “how unsatisfactory I for my part regard this position to be and how quixotic the path by which it has been arrived at”, and called mesothelioma claims “a category all their own” [S28]. - Insurance. Durham v BAI (“Trigger”) [2012] UKSC 14 held that employers’ liability policies respond, because the law “accepts a weak or broad causal link” [S29]. - Extension beyond mesothelioma. Heneghan v Manchester Dry Docks [2016] EWCA Civ 86 applied Fairchild to asbestos-induced lung cancer. Because s.3 covers only mesothelioma, Barker apportionment applied, and liability was several [S30]. - Foreseeability tightened, then loosened again. - Maguire v Harland & Wolff [2005] EWCA Civ 1 (appeal allowed, with a dissent) held that a shipyard could not reasonably have foreseen, in 1961–65, the risk to a wife washing her husband’s dusty clothes. The Court read Margereson as “fact-specific” and confined the “environmental principle” to conditions that “effectively replicated” factory exposure [S31]. - Williams v University of Birmingham [2011] EWCA Civ 1242 judged foreseeability “by the standard of the well-informed [defendant] … and not by the standard of omniscient hindsight” [S32]. - Bussey v Anglia Heating [2018] EWCA Civ 243 then corrected the reading of Williams as making pre-1970 exposure below the Technical Data Note 13 guidance levels “safe”. Underhill LJ framed the question as whether the employer should have been aware of “a significant risk of asbestos-related injury” (“any real risk, albeit statistically small, of a fatal illness is significant”) [S33].

The class-of-harm test survives for asbestos personal injury. Its application remains fact-specific and contested. - Administrative complements. The UK “2008 scheme” pays lump sums for mesothelioma, including environmental exposure such as having “lived near a factory using asbestos”. The Diffuse Mesothelioma Payment Scheme (DMPS; Mesothelioma Act 2014) pays victims who “cannot trace the employer … or their insurers”. It is funded by a levy on “active insurers” by market share [S34][S35]. This is administrative redress, not tort doctrine; see claim 6. - Environmental damage. I found no UK case applying Margereson-type class-of-harm foreseeability to environmental damage. Cambridge Water (1994), under which pollution is not actionable “because the type of harm was not reasonably foreseeable”, remains the leading authority. It is restated in Jalla v Shell [2023] UKSC 16, which also held that a one-off oil spill is not a “continuing nuisance” for limitation purposes, so the claimants lost on limitation [S36]. [Absence not verified by open search.] - A related success the chapter missed. Corby Group Litigation [2009] EWHC 1944 (TCC), decided 29 July 2009, found for the claimants on the generic issues. The substances released during a council’s reclamation of a steelworks site were capable of causing the limb defects, and there was an unexplained cluster [S37]. This is a notable counter-example to Graham v ReChem (p. 591), and it predates the chapter. - The EU direction runs partly the other way. The revised Product Liability Directive (EU) 2024/2853, to be transposed by 9 December 2026 [S38]: - retains the “development risk” defence (“objective state of scientific and technical knowledge … was not such that the defectiveness could be discovered”, Art. 11(1)(e)), which runs directly against Margereson-style reasoning, though member states may derogate (Art. 18); - adds rebuttable presumptions of defect and causation where a claimant “faces excessive difficulties, in particular due to technical or scientific complexity” and shows that causation is “likely” (Art. 10(4)); - extends the long-stop to 25 years for latent personal injury (Art. 17(2)).

Verdict: partly held up. - For personal injury the direction the panel identified continued, and it went further via statute (s.3), insurance law (Trigger) and extension to lung cancer (Heneghan). Courts nonetheless treated mesothelioma as exceptional, and foreseeability rulings cut both ways. - The suggested extension to environmental damage did not happen in UK law, on the evidence found.

Implications for weight. - The mechanisms “the first recognised harm is rarely the last” and “hindsight runs both ways” (Hale LJ) remain good law in the asbestos setting. - As a forecast of general long-tail liability, the panel overreached. The mesothelioma rules stayed special. - EU law instead chose to shift proof burdens through procedural presumptions rather than to relax foreseeability.

5. Early- or mid-life exposure to paraquat or TCE “may hasten the early onset of Parkinson’s disease, as shown by animal and human studies” (p. 586)#

Original claim (p. 586, citing Cranor 2011).

Subsequent developments: TCE - Epidemiology. Goldman et al., JAMA Neurology, July 2023, compared 172,128 Marines and Navy personnel at Camp Lejeune (water contaminated with TCE at monthly medians “greater than 70-fold the permissible amount”) with 168,361 at Camp Pendleton. Camp Lejeune veterans had “a 70% higher risk of PD (odds ratio, 1.70; 95% CI, 1.39-2.07)” and more prodromal signs [S39]. This builds on the 2012 twin study (Goldman et al., Ann Neurol) that predates the chapter [S40]. - Compensation. The VA lists Parkinson’s disease among eight presumptive conditions for 30 or more days at Camp Lejeune between 1953 and 1987 [S41]. The Justice Department’s Camp Lejeune “Elective Option” (6 September 2023) puts Parkinson’s disease in “Tier 2”, the group where the Agency for Toxic Substances and Disease Registry (ATSDR) found evidence “at an ‘equipoise and above’ level” [S42]. - Regulation. - US EPA issued a final Toxic Substances Control Act (TSCA) rule in December 2024 prohibiting most TCE uses. In September 2025 EPA extended some compliance dates, for example to 2028 for a nuclear-fuel processing use, and signalled reconsideration of parts of the rule [S43]. - The EU placed TCE on the REACH authorisation list in April 2013 (sunset date 21 April 2016), on carcinogenicity grounds [S44]. - The regulatory drivers were chiefly endpoints other than Parkinson’s disease.

Subsequent developments: paraquat - Epidemiology, supportive. A 2019 meta-analysis of 13 case-control studies found odds ratio 1.64 (95% CI 1.27–2.13) [S45]. - Epidemiology, against. The largest prospective cohort, the Agricultural Health Study (Shrestha et al. 2020; 38,274 applicators and 27,836 spouses), found associations for several other pesticides but not paraquat [S46]. EPA notes that this update “did not replicate earlier 2011 findings” [S47]. A 2021 single-author “review of reviews” by a private consultant (DLW Consulting) asserted that “a consensus exists … that the available evidence does not warrant a claim that paraquat causes Parkinson’s disease” [S48]. - Regulators. - EPA’s 2019 systematic review found “limited, but insufficient epidemiologic evidence”. Its current page (updated 13 November 2025) says EPA “has not found a clear link”. - In November 2025 EPA reported “greater uncertainty” about paraquat’s volatilisation and announced a data call-in [S47]. - Paraquat has not been approved in the EU since the General Court annulled its approval in 2007, before the chapter (Case T-229/04) [pre-2013 context; not re-verified in this pass]. - Occupational compensation for pesticides as a class. - France created agricultural occupational-disease table 58, “Maladie de Parkinson provoquée par les pesticides”, by decree of 4 May 2012. It requires 10 years’ exposure and covers diagnosis within 20 years [S49]. - Germany’s list of occupational diseases now includes BK 1322, “Parkinson-Syndrom durch langjährig, häufig und selbst angewendete Pestizide” (Parkinson’s syndrome from long-term, frequent, self-applied pesticides). It was added after 2017 [S50].

These presumptions concern pesticides generally and heavy occupational use, not paraquat specifically. - Litigation. - The paraquat MDL (over 5,000 plaintiffs) lost its sole general-causation expert on 17 April 2024 [S26]. - A Master Settlement Agreement was executed on 3 September 2025. It uses a “settlement matrix tier” for each eligible plaintiff, and opt-outs continue to be litigated into 2026 [S51]. - The terms of the settlement are not public on the court page.

Verdict: - TCE: strengthened. New large human evidence, and state recognition through compensation. - Paraquat: contested. Case-control meta-analyses show an association; the main cohort and the US regulator do not. Compensation exists for heavy occupational pesticide exposure as a class. - Early onset: unclear. The specific “early onset” formulation was not directly tested in the sources found. - The phrase “as shown by animal and human studies” overstated the paraquat evidence in 2013 and still does.

Implications for weight. The underlying mechanism has good support: long latency and subtle, rare outcomes make human evidence slow and insensitive. The single chemical examples should be cited with their different evidential status, and not as a pair.

6. No-fault administrative compensation for harms from emerging technologies and chemicals (pp. 581, 595–597, 601)#

Original claim. The chapter proposes: - presumptive “on-table” injury lists; - “the benefit of scientific doubt” for victims; - possibly the UK radiation scheme’s graded awards below 50% probability; - funding by levies rated on firms’ safety records (p. 603).

It is illustrated with the VICP, workers’ compensation, the 9/11 fund, the Compensation Scheme for Radiation-Linked Diseases (CSRLD) and New Zealand’s accident compensation scheme.

Subsequent developments - No general schemes. No no-fault scheme for harms from chemicals generally, GM crops, nanomaterials or mobile phones was found [absence not verified by open search]. For GMOs, the Commission’s 2016 evaluation of the Environmental Liability Directive (ELD) reported “no incidents of environmental damage caused by GMOs in the EU” in the reporting period. It left financial security voluntary while noting that “problems persist regarding the application of the Directive to large-scale accidents and insolvency” [S52]. - Presumptive schemes did grow in narrower forms. - Occupational pesticide diseases. France’s table 58 (2012) and the Fonds d’indemnisation des victimes de pesticides [S49][S53]. The fund was created by the 2020 social security financing law and has been fully operational since 30 November. It also covers children harmed by a parent’s workplace exposure before birth. Funding reportedly includes a levy on pesticide sales [not verified in this pass]. Germany added BK 1322 [S50]. - Pollution caused by government. The Camp Lejeune Justice Act 2022 (in the PACT Act, approved 10 August 2022) lets claimants prove a relationship “sufficient to conclude that a causal relationship is at least as likely as not” [S54]. This is a statutory “benefit of the doubt” at equipoise. It echoes the VA rule, in force since 1988, that “when there is an approximate balance of positive and negative evidence … the Secretary shall give the benefit of the doubt to the claimant” [S55].

The Elective Option grid pays USD 100,000–450,000 by tier and exposure duration, plus USD 100,000 for death. Tiers track ATSDR's grading of causal evidence: "sufficient" for Tier 1, "equipoise and above" for Tier 2 [S42].

As of 18 September 2026, offers exceeded USD 1.1 billion and payouts USD 879 million [S56]. This is for exposures that occurred 1953–87.

The Injury Table was substantially revised in January 2017 and amended in December 2021. The current table includes shoulder injury related to vaccine administration (SIRVA) and Guillain-Barré syndrome after seasonal influenza vaccine [S58]. - UK CSRLD. The 2024–25 Annual Statement (June 2025) reports [S59]: - “1823 cases considered/received since the Scheme began”, of which “202 … have resulted in successful claims” (about 11%), compared with 106 successes cited in the chapter; - 23 new claims in 2024/25, with “no claims … successful”; - the probability-of-causation schedules still date from 2009; - the 20% threshold remains. - Chapter errors relevant here. - The VICP figure of “USD 78 000 per plaintiff” should be about USD 787,000, and the program had run far longer than “12 years” by 2010. - The New Zealand account, that the scheme “reintroduced the notion of fault” for medical injury, appears outdated: the 2005 reform introduced no-fault “treatment injury” [not verified in this pass].

Verdict: partly held up. The preferred model, presumptive tables with graded or benefit-of-doubt standards, was adopted in several places. It happened only in settings with long exposure histories and many prior victims, and mostly for workers or for harm caused by government: - occupational pesticides after decades of epidemiology; - Camp Lejeune about 35 years after exposure ended; - asbestos.

This is the chapter’s own caveat (pp. 595, 599, 601) borne out. Nothing comparable exists for the emerging technologies it names. Levies rated on safety records were not a feature of the new schemes found.

Implications for weight. - The dilemma “tables require a history of prior harm” deserves strong weight. - The proposal to extend such schemes ahead of evidence to novel technologies has no real-world test. Claim 7 suggests a hazard for which pre-emptive tables would have been premature. - Administrative schemes can still be slow and can run on settlement (VICP, Camp Lejeune). “No-fault” does not mean “fast”.

7. Radiofrequency and brain cancer: evidence “may or may not be sufficient”, but compensation tables could be developed before causal evidence is sufficient (p. 595)#

Original claim (p. 595).

Subsequent developments - WHO-commissioned systematic review (Karipidis et al., Environment International, September 2024; 63 studies, published 1994–2022) [S60]. It found “moderate certainty evidence that it likely does not increase the risk of glioma, meningioma, acoustic neuroma, pituitary tumours, and salivary gland tumours in adults, or of paediatric brain tumours”. The meta-relative risk for glioma was 1.01 (0.89–1.13). Simulations of incidence trends showed relative risks above 1.5 with 10 years’ induction to be “definitely implausible”. The review was commissioned by WHO and part-funded by public bodies; one author is a member of the International Commission on Non-Ionizing Radiation Protection (ICNIRP). - Cohorts. The UK Million Women Study update (Schüz et al., JNCI, 2022) found relative risk 0.97 (0.90–1.04) for all brain tumours with ever-use [S61]. The IARC Advisory Group summarised MOBI-Kids (2022, no increased risk) and COSMOS (2024; hazard ratio 1.00 for glioma per 100 hours of calls) [S62]. - The question is not closed. IARC’s Advisory Group (report November 2024) rated RF-EMF “High priority (and ready for evaluation within 5 years)”. It called the human evidence “mixed” and noted “new evidence of carcinogenicity in experimental animals” (the NTP 2018 rat study), but said “a change in the current classification … is uncertain” [S62]. The 2011 Group 2B classification stands pending that evaluation.

Verdict: weakened as a hazard example; the proposal itself was untested. Cranor hedged (“may or may not”), but the example now cuts against the proposal. Tables built in 2013 “before the causal evidence is fully sufficient” would probably have compensated a harm that the best current evidence suggests does not exist at population level.

Implications for weight. This illustrates the chapter’s value choice (p. 581: “benefit of scientific doubt … given to victims”). Pre-emptive compensation moves the cost of false positives onto producers and levy-payers. That trade-off should be made explicitly and not presented as a technical fix. The mechanism is sound; this particular example lowers the weight of the proposal.

8. Establishing that BPA contributes to metabolic syndrome, breast cancer or reproductive effects is “extremely difficult”, unlike coke-oven emissions and lung cancer (“very easy”) (p. 595)#

Original claim (p. 595).

Subsequent developments - EFSA re-evaluation (EFSA Journal, 19 April 2023) [S63]: - identified immune effects (Th17 cells in mice) as the critical endpoint; - set a tolerable daily intake (TDI) of 0.2 ng/kg bw/day, down from a temporary TDI of 4 µg/kg bw/day in 2015, a 20,000-fold reduction; - found dietary exposure exceeding it “by two to three orders of magnitude”; - noted adverse metabolic, reproductive and developmental effects in a similar dose range, in animals. - Divergent scientific views. BfR (Germany’s Federal Institute for Risk Assessment, Opinion 018/2023, 19 April 2023) did “not support the new TDI”. It found “no evidence that the observed relative increases in the levels of Th17 cells trigger any adverse effects”, derived a TDI of 0.2 µg/kg bw/day (1,000 times EFSA’s), and noted that the European Medicines Agency (EMA) also presented different views [S64]. The US FDA states “BPA is safe at the current levels occurring in foods” (page content current as of 20 April 2023) [S65]. - EU restriction. Commission Regulation (EU) 2024/3190 of 19 December 2024 bans BPA, and hazard-classified bisphenols, in food-contact materials [S66]: - most single-use and repeat-use articles may be placed on the market until 20 July 2026; - some until 20 January 2028; - repeat-use articles until 20 January 2029 at the latest.

A 2026 amending regulation, 2026/250, concerns the transitional provisions [S67]. The restriction rests on EFSA’s TDI, not on proof of human disease. - Cancer evidence. IARC’s Advisory Group (2024) rated BPA high priority [S62]. It reported that meta-analyses from 2021 and 2023 “did not support an association between BPA and breast cancer”. It found some weaker prostate-cancer signals, and tumours in rats in a well-conducted study following good laboratory practice (GLP).

Verdict: held up. Thirteen years later, proof that BPA causes human breast cancer or metabolic disease remains elusive and contested. The EU nevertheless restricted it, on hazard and precautionary grounds. That supports the chapter’s broader point that proof requirements for compensation and for regulation differ.

Implications for weight. This strongly supports the lesson that compensation tables are hardest exactly where exposures are ubiquitous, low-dose and act through endocrine or immune pathways. It also shows the complementarity the chapter assumes (p. 582): regulation acting ex ante can move where compensation cannot.

9. Assurance bonds sized to worst-case damages, with the burden on the firm (“no bond, no market”); Deepwater Horizon damages estimated at USD 34–670 billion against a USD 50 million cap; a hypothetical USD 50 billion bond (pp. 600–603, Panel 24.3)#

Original claim (Panel 24.3, from Costanza et al. 2010; pp. 600–601).

Subsequent developments - Actual costs. BP’s 2018 Annual Report and Form 20-F (filed 29 March 2019) reports a “cumulative pre-tax income statement charge since the incident” of USD 67.0 billion, with payments expected of “around $1 billion per year for the next 14 years” [S68]. This lies at the low end of Costanza’s range and excludes most non-market ecosystem losses. - Key legal steps (EPA chronology) [S69]: - USD 4 billion BP criminal plea, 15 November 2012; - Transocean USD 1 billion civil settlement and USD 400 million criminal plea, 3 January 2013; - Phase One findings of “Gross Negligence and Willful Misconduct”, 4 September 2014; - USD 14.9 billion civil settlement with BP, 5 October 2015, including “an unprecedented $5.5 billion Clean Water Act penalty and up to $8.8 billion in natural resource damages”. - The cap. The chapter’s figure is wrong. Under the Oil Pollution Act (OPA), the cap for an offshore facility is “the total of all removal costs plus $75,000,000”. It does not apply where the incident was caused by “gross negligence or willful misconduct” or by violation of federal safety regulations [S70]. In the event, BP’s payments dwarfed the cap.

After the spill, Congress did not raise the statutory cap [not verified in this pass]. BOEM instead indexed it to inflation by regulation from 2014. It now stands at “the total of all removal costs plus $167.8069 million” (30 CFR 553.702, as amended 14 April 2023) [S71]. Required oil-spill financial responsibility (OSFR) still tops out at USD 150 million, and “the Director will not require an OSFR demonstration that exceeds $150 million” (30 CFR 553.13) [S71]. - The closest adoption of “no bond, no market”. The EU Offshore Safety Directive 2013/30/EU (12 June 2013) requires that a licence not be granted “unless [the authority] is satisfied … that the applicant has made or will make adequate provision … to cover liabilities potentially deriving from the applicant’s offshore oil and gas operations” (Art. 4(3)) [S72]. It does not use worst-case sizing, reversed burden or refundable bonds; the arrangements are left to member states. - Decommissioning. BOEM strengthened supplemental financial assurance for decommissioning liabilities in 2024 (30 CFR 556.901, as amended 24 April 2024) [S73]. This addresses the “marginal firm” problem (p. 601) for known liabilities, not uncertain damages. - ELD. Financial security remains voluntary (2016 evaluation) [S52]. - Nuclear liability. Caps were raised, not removed. The UK’s 2016 Order implementing the 2004 Paris/Brussels Protocols sets operator liability at EUR 700 million, “rising incrementally to €1200 million over 5 years” [S74]. - Novel hazards. No example was found of Costanza-style worst-case assurance bonds, with a reversed burden, adopted for novel technological hazards [absence not verified by open search].

Verdict: partly held up. - Damage estimate: held up. The realised private cost of USD 67 billion falls inside the 34–670 billion range. - Critique of liability caps: held up in principle but misstated in detail. The cap was USD 75 million plus uncapped removal costs, and it proved irrelevant because of gross negligence and BP’s ability to pay. - The bond proposal: not adopted. Instruments moved only toward licence-stage financial-capacity checks (EU) and decommissioning assurance (US).

Implications for weight. - The mechanism that caps combined with a public burden of proof encourage excessive risk-taking has moderate support. - Deepwater Horizon is actually a weak test of it: the firm was solvent and the cap fell away. - The case for bonding rests more strongly on the “judgment-proof marginal firm” argument (p. 601) than on this example. - Treat “no bond, no market” as an untested design idea.

10. “Some evidence that young scientists are being discouraged from entering controversial fields for fear of such harassment” (summary box, p. 581)#

Original claim (p. 581). It is unsupported in the chapter body.

Subsequent developments - Evidence available then (not cited by the chapter): - Kempner (PLoS Medicine, November 2008) studied NIH grantees whose grants were attacked politically (30 interviews, 82 surveys). “About half” now removed controversial words from grants, “a quarter reported eliminating entire topics from their research agendas”, and “four researchers reportedly chose to move into more secure positions entirely” [S75]. - An earlier Science survey (Kempner et al. 2005, “Forbidden knowledge”) documented informal constraints on research topics [S76].

This supports a chilling effect on established researchers’ choice of topics. It does not directly address young scientists’ choice of field. - Since: - Surveys of harassment during the COVID-19 pandemic (Nature, October 2021; Science, March 2022) documented abuse and threats against scientists who spoke publicly [S77][S78]. Survey figures were not retrieved in this pass. - The ALLEA code (2023) explicitly states that “early career and short-term employed researchers may be particularly vulnerable” [S19]. - The Maddox Prize created an early-career category [S18]. - The Commission’s 2024 report refers to a “chilling effect on potential whistleblowers” where protection is lacking, a general statement [S2]. - Not found. No study directly measuring whether young scientists avoid controversial fields because of fear of harassment was located [not verified by open search].

Verdict: unclear. The broad claim of a chilling effect on research agendas has empirical support, some of it predating the chapter. The specific claim about young scientists’ entry into fields remains unsupported by direct evidence in the sources found. It is plausible and recognised by institutions, but not demonstrated.

Implications for weight. Cite the chilling-effect mechanism (Kempner 2008) rather than the chapter’s sentence. Treat the claim about young scientists as a hypothesis.

Cross-cutting observations#

  1. Mechanisms, not proposals, carry the weight. Post-2013 events repeatedly confirm the chapter’s structural points: - proof rules allocate the cost of uncertainty: Milward, the paraquat MDL, Rule 702, Sienkiewicz; - tables need earlier victims: Camp Lejeune, the pesticide tables, the stalled CSRLD; - latent harm meets vanishing defendants: DMPS, BOEM decommissioning assurance; - regulation can act where compensation cannot: BPA.

The chapter’s reform proposals were largely not adopted, or were adopted only in narrow, backward-looking forms. 2. Reversibility. The closest institutional embodiment of “protecting early warners” (France’s cnDAspe, 2013–2026) was abolished in a simplification law. Protections for warners are politically fragile, even where adopted. 3. Selection effects confirmed. The one existing prize model honoured a retracted study (2015). The RF example has weakened. Both illustrate the false-positive cost that the chapter and Panel 24.1 acknowledge but play down. 4. Accuracy and disclosure. - The Deepwater Horizon cap is misstated (USD 50 million instead of USD 75 million plus removal costs, subject to exceptions). - The VICP average is off by a factor of about 10. - The New Zealand account appears outdated [not verified]. - Cranor’s expert role in Milward is not disclosed. - None of these overturns a structural point, but together they argue for using the chapter as a source of mechanisms and designs, not of facts. 5. Where the law actually moved. EU law chose rebuttable procedural presumptions in the revised Product Liability Directive (2024) and a reversed burden on retaliation in the Whistleblowing Directive (2019), and kept the development-risk defence. Burden shifting has proved more politically acceptable than relaxing foreseeability or introducing no-fault funds.

Sources#

All accessed 26 September 2026 unless stated. Dates are those of the document.