Late Lessons, Jensen Huang and AI
Markdown version

LL2-A3 hindsight check: Annex 3, “An update of some case studies from vol. 1”#

EEA Report No 1/2013, report pp. 717–737 (PDF 719–739). Hindsight window: 2013 to late September 2026. Working files: digest digests/LL2-A3.md; source text text/chunks/LL2-A3.txt.

Overview#

Annex 3 is already a hindsight exercise. It has nine short updates, written in 2012–13, of 2001 case studies. This check asks the same question again, thirteen years on. Two things matter for how much weight the annex can carry.

1. The direction of its warnings has mostly held, and several have been formally vindicated. - IARC classified automotive gasoline as a Group 1 human carcinogen in 2025. - Estimates of the global asbestos death toll have roughly doubled, and the EU cut its occupational asbestos limit by 10–50 times, moving to electron microscopy. - EFSA lowered its tolerable intake for BPA 20,000-fold, and the EU banned BPA in food-contact materials. - The EU, US and China ended antimicrobial growth promoters, and the EU extends that requirement to imports from 3 September 2026. - The Montreal Protocol has so far delivered the recovery path the annex described.

2. Many of the specific quantitative, mechanistic or causal claims have not been confirmed. Several have been scaled back. These are mostly the claims that rested on the contributors’ own, unpublished or uncited work: - Farmed-fish pollutants “fully abolishing” the benefits of fish. - An epigenetic pathway from PCBs to diabetes, with “additive” effects on 14 genes. - MTBE as a “probable” human carcinogen, based on lymphomas in one laboratory’s rats. - Benzene risk underestimated 3–9-fold. - Missing leukaemia deaths in the Pliofilm cohort.

In short, the annex is a good guide to which way hazards and regulation later moved. It is a weak guide to how large the effects are or by what mechanism they act.

Two scope points: - This file tests the annex’s own claims. The corresponding 2001 chapters were checked separately. - Method caveat: the web-search budget was exhausted early in this run. The check therefore relies on primary documents fetched directly: EU law via the Publications Office, IARC, USGS, HSE, WTO, the Rotterdam Convention secretariat, WHO, FDA, EMA, and peer-reviewed abstracts via Europe PMC. Three items are given from memory and marked [not re-verified].

Verdicts at a glance#

# Claim (page) Verdict
1 Harm “expands over time”, often at exposures once thought safe (p. 717) Held up (with a selection caveat)
2 Pollutants in farmed salmon “can fully abolish” fish benefits; regulatory gaps (p. 718) Partly held up
3 Benzene/gasoline cause the full range of blood cancers below 1 ppm; risk underestimated 3–9× (pp. 720–721) Partly held up (direction strengthened, breadth and magnitude overstated)
4 Asbestos: ≥107,000 deaths a year; bans offset by Asian growth; road ahead difficult (pp. 724–726) Held up (burden strengthened; “offset” and “stable 2 Mt” overtaken)
5 PCBs cause diabetes/obesity via prenatal epigenetics; “denial” blocks innovation (p. 727) Contested
6 Ozone recovers around mid-century, “assuming global compliance” (p. 728) Held up
7 DES harms keep widening, with third-generation effects; BPA is a DES-type warning (pp. 729–730) Strengthened (DES third-generation signals modest; BPA still contested outside the EU)
8 Growth-promoting antimicrobials have little benefit; bans cost nothing; use will grow globally (pp. 731–733) Held up (policy vindicated; the “net loss” economics not generalised)
9 MTBE is carcinogenic in rodents and “probably” in humans; children at risk (p. 735) Partly held up (animal carcinogenicity confirmed on other tumours; human “probable” not supported)
10 2008 WTO Appellate Body report is a “significant victory” that settles the law for the EU (pp. 736–737) Partly held up

Claim 1: The editors’ thesis that harm expands over time (p. 717)#

Original claim. “For most of the harmful agents described in the case studies, their capacity to harm expands over time from that which generated the first ‘early warnings’ to other kinds of harm and which is often caused by levels of exposure that were previously considered ‘safe’” (p. 717).

Subsequent developments. Across the annex’s own agents, the post-2013 record mostly fits the pattern.

Where harm widened: - Asbestos. Occupational death estimates rose from about 107,000 to more than 200,000 a year (WHO fact sheet, 11 Aug 2026). IARC’s site list now includes larynx and ovary as sufficient-evidence sites (IARC list by cancer site, updated 18 May 2026). - Benzene and gasoline. IARC added limited-evidence sites for benzene in 2017, and in 2025 classified gasoline Group 1 with bladder cancer as a new sufficient-evidence site (claim 3). - PCBs. IARC upgraded PCBs to Group 1 (melanoma) in 2013. EFSA cut the tolerable weekly intake for dioxins and dioxin-like PCBs seven-fold in 2018. - BPA. EFSA cut its tolerable daily intake 20,000-fold in 2023, identifying immune effects as the most sensitive endpoint (claim 7). - DES. Third-generation signals appeared (claim 7).

Where it did not: - MTBE. The claimed lymphoma/leukaemia and human-carcinogen profile was not confirmed (claim 9). - Farmed fish. Risk assessors continue to judge the benefits to outweigh the risks (claim 2). - DES sons. No excess cancer was found in DES-exposed men, and there was no overall cancer excess in DES daughters (claim 7). - Hormone residues. No new human harm evidence surfaced (claim 10). - Ozone. The ozone case is a story of harm contained, not expanded (claim 6).

Falling “safe” levels: - The EU benzene limit fell from 1 ppm to 0.2 ppm (Directive (EU) 2022/431). - The EU asbestos limit fell from 0.1 to 0.01 f/cm³, with 0.002 f/cm³ (or electron-microscope counting of thin fibres) from December 2029 (Directive (EU) 2023/2668). - The dioxin and BPA intake values were cut, as above.

Verdict: held up, with a selection caveat. The cases were chosen because harm occurred, and continued study of a known hazard tends to find more endpoints. The pattern is real but partly produced by where the research goes. The counter-cases (MTBE, fish, hormone residues) show that expanding harm is common but not automatic.

Implications. “Harm profiles widen and thresholds fall with continued study” (digest insight 1) can be used as a strong prior for agents already shown to be harmful. It should not be used to presume harm for a newly flagged agent. The evidence fits a narrower mechanism: once a hazard is established, more sensitive endpoints and better instruments tend to lower the level of concern.

Claim 2: Ruzzin on farmed fish (p. 718)#

Original claim. - In the author’s rodent studies, background persistent organic pollutants (POPs) in farmed Atlantic salmon “can fully abolish the potential health benefits of fish nutrients”. - The EPA+DHA to n-6 ratio in farmed salmon fell about 50% in 2005–09. The consequences were unstudied but “likely” reduced the benefits. - Organochlorine pesticides, some PCBs, PBDEs and PFCs were unregulated in EU seafood. - The health benefits of fish “may be over-estimated”. There is an “urgent need” to regulate and study (p. 718).

Subsequent developments. - Omega-3 decline: confirmed. In over 3,000 Scottish farmed salmon from 2006 to 2015, EPA+DHA fell as terrestrial feed ingredients rose. Double portions were needed by 2015 to meet recommended intakes, though farmed salmon still delivered more EPA+DHA than most fish (Sprague, Dick & Tocher, Sci Rep, 22 Feb 2016). - Pollutant levels in farmed salmon: fell. Monitoring of more than 2,300 Norwegian farmed salmon found mercury, arsenic, dioxins, dioxin-like PCBs and DDT decreased between 1999 and 2011 (Nøstbakken et al., Environ Int, Nov 2014). Ruzzin and colleagues published a critical comment and the authors replied (Environ Int, 2015). The disagreement was about interpretation against tolerable intakes, not the downward trend. The shift to plant-based feed that lowered the omega-3s also lowered marine-derived POPs, a trade-off the annex does not mention. - “Background levels are not safe”: partly supported by regulators. - EFSA’s 2018 opinion cut the tolerable weekly intake for dioxins and dioxin-like PCBs to 2 pg TEQ/kg bw/week. It concluded that mean and high-end European intakes “considerably” exceed it (EFSA CONTAM, 20 Nov 2018). - EFSA’s 2020 opinion set a group tolerable weekly intake of 4.4 ng/kg bw/week for four PFAS. - Regulatory gaps: partly closed. - EU maximum levels for PFOS, PFOA, PFNA and PFHxS in fish, eggs and meat have applied since 1 January 2023 (Regulation (EU) 2022/2388, now consolidated in Regulation (EU) 2023/915). The “PFCs unregulated” gap is closed for these four substances. - Non-dioxin-like PCBs (the six indicator congeners) already had maximum levels from 2012, so the annex’s “some PCBs” was already narrowing when it was written. - The consolidated contaminants regulation still contains no maximum levels for PBDEs. - “Fully abolish the benefits”: not adopted by risk assessors. - EFSA’s 2014 opinion on seafood benefits versus methylmercury concluded that about 1–2 servings a week (3–4 in pregnancy) are associated with better neurodevelopmental and coronary outcomes. It found benefits outweigh methylmercury risk for most species (EFSA NDA, July 2014) [summary not re-verified; the abstract could not be retrieved]. - I found no later authoritative benefit–risk assessment that endorses the rodent-based conclusion. - Fish and type 2 diabetes: the alarm faded. - A 2017 meta-analysis of prospective studies found no significant association between fish intake and type 2 diabetes, with the evidence graded “moderate” (Schwingshackl et al., Eur J Epidemiol, Apr 2017). - A 2019 BMJ meta-analysis of 83 randomised trials found long-chain omega-3 had “little or no effect” on diabetes diagnosis (RR 1.00, 95% CI 0.85–1.17) (Brown et al., 21 Aug 2019). - Omega-3 and the heart: the picture is mixed. The VITAL trial found no reduction in major cardiovascular events from marine n-3 supplements (NEJM, Nov 2018). The 2020 Cochrane review found long-chain omega-3 “slightly reduces” coronary heart disease mortality and events (Abdelhamid et al., 29 Feb 2020). Both concern supplements, not contaminated fish.

Verdict: partly held up. - The factual observations (the omega-3 decline, the regulatory gaps) were accurate and have been corroborated or acted on. - The alarming causal claim (background POPs “fully abolish” benefits) rested on the author’s rodent work. It has not become a mainstream conclusion.

Implications. - The annex’s insight that “systems optimised for producer metrics degrade properties that matter to users” is supported by the feed-substitution evidence (Sprague 2016). - But the lesson must carry its trade-off: the same substitution reduced one hazard while reducing a benefit. - “Background exposure may be less safe than assumed” is supported for dioxins and PFAS through lowered tolerable intakes, not through the rodent studies. - Treat the fish update as a correct early warning about nutritional drift and regulatory gaps, and as an unconfirmed warning about net health harm.

Claim 3: Infante on benzene and gasoline (pp. 720–721)#

Original claim. - Benzene raises the risk of “all of the major forms of leukemia” (AML, ALL, CML, CLL), plus non-Hodgkin lymphoma and multiple myeloma, at average exposures below 1 ppm. - Gasoline “should be considered a cause of all lymphohematopoietic cancers associated with benzene exposure to both adults and children”. - Quantitative risk assessments underestimate environmental risk 3–9-fold because metabolism is more efficient below 1 ppm. - The Pliofilm cohort omits about 4–9 leukaemia deaths. This last point is cited to two papers “submitted for publication” (pp. 720–722).

Subsequent developments. - IARC on benzene (Monograph 120; meeting 2017, published 2018). Group 1, with sufficient evidence for AML in adults. The IARC site list shows limited evidence for CML, CLL, non-Hodgkin lymphoma, multiple myeloma, childhood AML and lung cancer. Adult ALL is not listed even as limited (IARC list by cancer site, 18 May 2026; Loomis et al., Lancet Oncol, Dec 2017). So IARC saw positive associations for most of Infante’s list, but sufficient evidence only for AML. - IARC on gasoline (Monograph 138; evaluation published 21 March 2025; volume 2026). Automotive gasoline was upgraded to Group 1. There was sufficient evidence in humans for AML and bladder cancer, and limited evidence for childhood ALL, non-Hodgkin lymphoma/CLL, multiple myeloma, myelodysplastic syndromes, and stomach and kidney cancers (IARC news, 21 Mar 2025; Turner et al., Lancet Oncol 2025). This strongly vindicates the thrust of Infante’s gasoline claim, including the paediatric dimension, though not the claim of “all” such cancers at the sufficient level. - Low-level exposure: split evidence. - Independent study supporting an effect. A Norwegian offshore cohort with average intensities of about 0.04 ppm found dose-related patterns for AML and multiple myeloma, and suggestively for CLL (Stenehjem et al., Br J Cancer, 2015). - Industry-affiliated studies against. Pooled petroleum case-control studies (industry-affiliated authors) found the data “do not persuasively demonstrate” an AML risk at low exposures, with MDS the more relevant risk (Rushton et al., Br J Cancer, Dec 2013). They found no convincing CML association (Glass et al., OEM, Feb 2014). - Chinese cohort. A large Chinese case-cohort found the strongest effects for MDS/AML arising within 10 years of exposure and for exposure before age 30 (Linet et al., JNCI, May 2019). - Nonlinear metabolism: contested, openly. The Kim/Rappaport finding of more efficient metabolism below 1 ppm was reanalysed in an American Petroleum Institute-funded study (Price et al.). Rappaport et al. rebutted it as “scientifically unsound” (“Low-dose metabolism of benzene in humans: science and obfuscation”, Carcinogenesis, Dec 2012), and Price et al. replied (2013). I found no regulatory adoption of a 3–9-fold correction. - Childhood leukaemia. Meta-analyses supported associations: - For occupational or household exposure to benzene and solvents (parental, in utero or early life), summary relative risk 1.96; for traffic metrics, 1.48 (Carlos-Wallace et al., AJE, 2016). - An approximately linear dose–response for benzene and childhood leukaemia, especially AML (Filippini et al., EHP, Apr 2019). - Pliofilm. Infante published “Pliofilm revisited I” in 2013. It argues that 2–5 AML cases could be added to the Akron portion and documents company concealment (IJOEH, Jul 2013). The published “Part II” is about take-home leukaemia, not the statistical estimate cited in the annex. I found no independent reanalysis adopting the 4–9 missing deaths. - Regulation. - The EU binding occupational limit fell from 1 ppm to 0.2 ppm, with transitional values of 1 ppm to 5 April 2024 and 0.5 ppm to 5 April 2026. Recital 26 commits the Commission to assess “a further reduction” in light of the 2018 RAC opinion (Directive (EU) 2022/431, 9 Mar 2022). The RAC opinion recommended about 0.05 ppm [not re-verified]. - The EU ambient limit tightens from 5 to 3.4 µg/m³ by 2030 (Directive (EU) 2024/2881). - The US OSHA permissible exposure limit remains 1 ppm (29 CFR 1910.1028).

Verdict: partly held up. - The direction was strongly vindicated: gasoline is Group 1, the EU limit is lower, and low-level and childhood associations have accumulated. - The breadth (“all” major leukaemias) exceeds IARC’s sufficient evidence. - The magnitude claim (3–9-fold) and the Pliofilm correction remain contested or unconfirmed.

Implications. - The lesson “linear extrapolation is not automatically conservative” (digest insight 2) stays contested. The benzene debate is a live example of a funder-linked reanalysis contesting a low-dose finding, which supports the annex’s point about interests shaping evidence. - “One canonical dataset can anchor regulation for decades” (insight 3) is plausible. But the specific Pliofilm correction rests on one author’s historical reconstruction. - The strongest transferable point is procedural: an exposure source (gasoline) was eventually classified in its own right, decades after its main hazardous component (benzene).

Claim 4: Castleman and Gee on asbestos (pp. 724–726)#

Original claim. - At least 107,000 occupational asbestos deaths a year (WHO), with an eventual toll of 5–10 million (unsourced). - Consumption stable at about 2 Mt/yr since 2000, with bans in over 50 countries “offset” by growth in China, India and others. - Ottawa will no longer fight Rotterdam listing of chrysotile. - The European Parliament (13 March 2013) called for removal from public buildings by 2028. - Cheap phase-contrast microscopy shaped evidence and limits. - The road ahead is “difficult”, especially in India, China and Russia. - Peto: school-related mesothelioma deaths in women could fall tenfold in 50 years (pp. 724–726).

Subsequent developments. - Burden: revised sharply upward. - A 2018 synthesis estimated 255,000 asbestos deaths a year, 233,000 of them work-related (Furuya et al., IJERPH, 16 May 2018). - WHO now states “more than 200 000” occupational deaths a year, over 70% of all work-related cancer deaths (WHO fact sheet, 11 Aug 2026). - The Commission put EU-27 occupational asbestos deaths at over 70,000 in 2019 (COM(2022) 488, 28 Sep 2022). - The unsourced 5–10 million traces to LaDou’s estimate of “as many as 10 million” (EHP, 2004). Current annual estimates make it plausible over the coming decades. - Consumption: fell, not stable. USGS estimates world consumption at about 930,000 t in 2025, “a decrease of nearly 55%” from about 2 Mt in 2000. Mine output in 2025 was: Russia 310,000 t, China 250,000 t, Kazakhstan 250,000 t, Brazil 150,000 t (USGS Mineral Commodity Summaries, Feb 2026). Bans have not merely been “offset”. Global use has halved, though USGS expects demand to continue, particularly for construction in Asia. - National bans. - Canada banned asbestos from 30 December 2018 (SOR/2018-196). - The US EPA banned ongoing chrysotile uses in March 2024 (89 FR 21970, 28 Mar 2024), with imports banned from May 2024 and chlor-alkali diaphragms phased out by 2029–2036. USGS notes the dates “may be modified” by legal proceedings in 2025 (Feb 2026). - Brazil’s Supreme Court upheld in February 2023 its 2017 ruling that asbestos extraction, sale and use are unconstitutional. Yet Brazil mined about 150,000 t in 2025, largely for export under a Goiás state law, and a 2024 state phase-out clock had not started by September 2025 (USGS, Feb 2026). - Rotterdam Convention. Chrysotile remains “recommended for listing but not yet listed”. COP-12 (2025) listed carbosulfan and fenthion, not chrysotile (Rotterdam Convention secretariat pages, accessed Sep 2026). Canada’s withdrawal did not unlock listing, because other Parties continued to block it. - European Parliament’s 2028 target. I found no binding EU instrument adopting it. What was adopted: - Directive (EU) 2023/2668 (22 Nov 2023) cut the worker limit tenfold to 0.01 f/cm³. From 21 December 2029 member states must apply either 0.01 f/cm³ counting thin fibres by electron microscopy or 0.002 f/cm³. The Directive states that optical microscopy “does not allow the thinnest fibres detrimental to health to be counted”. This is a direct vindication of the annex’s point about cheap instruments shaping limits (p. 725). - The recast Energy Performance of Buildings Directive requires member states to “address” asbestos removal in buildings undergoing major renovation (Directive (EU) 2024/1275). This is a renovation-linked approach, not a deadline. - UK trend and Peto’s projection. GB mesothelioma deaths were 2,146 in 2024, below the 2,508 average for 2011–2020. Female deaths (375) are expected to decline only towards the end of the 2020s. Teachers show higher proportional mortality than several other non-asbestos occupations (HSE, July 2026). Peto’s 50-year projection cannot yet be tested.

Verdict: held up. - The burden estimate was conservative and has been roughly doubled. - The “difficult road” in producer and consumer states is borne out: Russia, Kazakhstan and China still mine, Brazil exports despite its own constitutional ruling, and Rotterdam listing is still blocked. - The claims of stable consumption at 2 Mt, and of bans merely “offset” by Asian growth, were overtaken by a halving of global use. - The 2028 removal target was not adopted.

Implications. This is the annex’s best-supported case for three lessons: - Displacement. Local restriction displaces the hazard to less-regulated markets, and producer states block information-sharing treaties (insight 8). Brazil adds a new example: a national prohibition coexisting with export mining. - Legacy. Relief from legacy harm takes decades (insight 12). - Measurement. Measurement technology shapes permissible limits (insight 4), now written into EU law.

The post-2013 record also shows that sustained coalitions can shrink global use substantially even without treaty listing. That point cuts against the annex’s more pessimistic framing.

Claim 5: Koppe on PCBs (p. 727)#

Original claim. - PCBs contribute to type 2 diabetes and obesity, with prenatal epigenetic changes as the key mechanism. - In Slovak children, PCB and DDT/DDE/HCB exposures changed the same 14 genes in the same direction, indicating “additive” effects. - “Denying toxic effects of PCBs … has hampered the innovative development to cure the epigenetic changes”, and “this denial is still” happening (p. 727).

Subsequent developments. - PCB hazard: strengthened. - IARC classified PCBs (and dioxin-like PCBs) as Group 1 in 2013 (sufficient evidence for melanoma; limited for non-Hodgkin lymphoma and breast cancer) (Lauby-Secretan et al., Lancet Oncol, 15 Mar 2013; IARC site list, May 2026). - EFSA’s 2018 opinion cut the tolerable weekly intake for dioxins and dioxin-like PCBs seven-fold (EFSA, 20 Nov 2018). - Diabetes: association yes, causation not established. - A US National Toxicology Program workshop concluded that evidence “is sufficient for a positive association of some organochlorine POPs with type 2 diabetes” but “not sufficient to establish causality” (Taylor et al., EHP, May 2013). - Meta-analyses report pooled relative risks around 1.7–2.4 for total PCBs, from cross-sectional and prospective data (Wu et al., EHP, 2013; Song et al., J Diabetes, 2016). - A GRADE-based review judged the evidence moderate for p,p′-DDE. For PCBs it was “poorer … it is likely that the rodent models used are not appropriate” (Lind & Lind, Diabetologia, May 2018). - Reverse causation and confounding by adiposity remain standard caveats. - Epigenetic mechanism and the 14 genes. I found no subsequent replication establishing the Slovak 14-gene “additive” finding, or a prenatal epigenetic pathway, as mainstream. - “Denial” and legacy management. - Continuing inaction is documented, as distinct from denial of toxicity. At most 30% of countries are on track for environmentally sound management of PCBs by the Stockholm Convention’s 2028 goal. Over 10 million tonnes of PCB-containing materials remain. The US, not a Party, reduced its PCB mass by only about 3% since 2006 (Melymuk et al., ES&T, Jun 2022). - The 2025 goal (remove PCB equipment from use) and the 2028 goal remain on the Convention’s books (Stockholm Convention PCB page, accessed Sep 2026). - No evidence emerged that “denial” blocked a cure for epigenetic changes. The claim is not testable as stated.

Verdict: contested. PCB toxicity in general was strengthened. The annex’s specific claims were not established: diabetes through epigenetics, additivity on 14 genes, and denial inhibiting cures.

Implications. Use this update only for the broad point that legacy persistent chemicals continue to reveal new endpoints and remain poorly managed. The Melymuk data give a strong, quantified example of “legacy stocks and long latency delay relief” and of implementation gaps under a treaty. The PCB update’s mechanistic claims should not be used as evidence for any lesson.

Claim 6: Ozone layer (p. 728)#

Original claim. This is a summary of the WMO/UNEP Twenty Questions: 2006 Update: - Ozone-depleting gases are declining under the Montreal Protocol. - Halons and HCFCs are still rising but will decline “if compliance … continues”. - Effective ozone-depleting gases fall to pre-ozone-hole values around mid-century, with “substantial recovery” near mid-century “assuming global compliance”. - Volcanic eruptions and climate change could delay recovery. - Substitute gases contribute to climate change (p. 728).

Subsequent developments. - Recovery dates. The 2022 WMO/UNEP Scientific Assessment projects total ozone returning to 1980 values around 2040 (near-global), 2045 (Arctic) and 2066 (Antarctic). The Antarctic range is 2049–2077 (Executive Summary, 2022/23). - Compliance tested. - Unexpected CFC-11 emissions after 2012 were traced largely to eastern China (Rigby et al., Nature, May 2019). - They then fell sharply in 2018–19 (Montzka et al., Nature, Feb 2021). - The 2022 Assessment states that uncertainties “are too large to determine whether all unexpected emissions have ceased”. It also flags unexplained emissions of several other CFCs, of carbon tetrachloride, and of HFC-23 (up to eight times higher than expected). - HCFCs. Equivalent effective chlorine and radiative forcing from HCFCs have declined since 2021, five years earlier than projected (Western et al., Nature Climate Change, 11 Jun 2024). - Substitutes and climate. The 2016 Kigali Amendment addresses the climate effect of HFC substitutes, which the annex noted but left unresolved. Compliance is estimated to avoid 0.3–0.5 °C of warming by 2100 (2022 Assessment). - Natural disruptions. Exceptional events of the kind the annex anticipated occurred: the 2019–20 Australian wildfires and the 2022 Hunga Tonga eruption. Their ozone effects were still under study (2022 Assessment).

Verdict: held up. The projections are on track. The “assuming global compliance” caveat proved to be the right one: it was breached and then largely repaired, detected by atmospheric monitoring.

Implications. This remains the annex’s one success story, and the editors never reconcile it with their thesis of expanding harm (digest caveat). Post-2013 events add three transferable mechanisms: - Treaty success depends on independent monitoring that can detect non-compliance (the CFC-11 episode). - Substitutes can carry a different harm that needs its own agreement (Kigali). - Recovery is measured in decades (insight 12).

Claim 7: Swan on DES, and the BPA warning (pp. 729–730)#

Original claim. - DES harms are far more extensive than predicted in 2000: 12 outcomes in daughters (hazard ratios 1.4–8.1) with dose gradients, and harm in sons. - Third-generation effects are “starting” to emerge. - BPA, recognised as oestrogenic in the 1930s, is now made at about 10 billion lb a year. Exposure in pregnancy is ubiquitous, and effect studies are “only now being conducted” (pp. 729–730).

Subsequent developments: DES. - Third generation. - DES granddaughters showed more irregular menses (prevalence ratio 1.32) and amenorrhea (1.26), and a higher risk of preterm delivery (RR 1.54). The associations were stronger when their mothers had vaginal epithelial changes (Titus et al., Reprod Toxicol, Dec 2018; NCI Third Generation Study). - In the Nurses’ Health Study II, grandmaternal DES use was associated with ADHD in grandchildren, with an odds ratio of 1.36, or 1.63 for first-trimester use (Kioumourtzoglou et al., JAMA Pediatrics, Jul 2018; self-reported exposure). - Cancer: follow-up tempered some fears. - In the combined NCI cohorts, clear-cell adenocarcinoma remained greatly elevated (SIR 27.6) and breast cancer modestly so. There was no overall cancer excess and no “diathesis of cancers” in DES daughters (Troisi et al., Environ Mol Mutagen, 2017/2019). - DES-exposed men showed no increase in overall or prostate cancer (Strohsnitter et al., CEBP, Jul 2021). - Benign breast disease was not elevated in DES daughters (Mitra et al., Int J Cancer, Jan 2026).

Subsequent developments: BPA. - EFSA 2023 re-evaluation. The TDI fell from 4 µg/kg bw/day to 0.2 ng/kg bw/day (20,000-fold), with Th17-cell immune effects as the critical endpoint. Dietary exposure in all age groups exceeds the TDI “by two to three orders of magnitude” (EFSA CEP, 19 Apr 2023). - Dissent. EFSA published joint reports on its methodological differences with the German BfR and the EMA (EFSA news, 19 Apr 2023). BfR proposed a TDI about 1,000 times higher [value not re-verified]. - EU ban. Commission Regulation (EU) 2024/3190 (19 Dec 2024) prohibits BPA and other hazardous bisphenols in food-contact materials, with transitions running to 2026–2029. - US position. The US FDA still states “BPA is safe at the current levels occurring in foods” (FDA BPA page, updated 20 Apr 2023). This leaves a clear transatlantic divergence.

Verdict: strengthened. - The DES record continues to widen, but modestly and unevenly. The third-generation effects are real signals but small, and cancer in sons and overall cancer in daughters are not elevated. - The BPA warning was substantially vindicated in EU regulation, though the scientific and regulatory dispute continues.

Implications. - DES supports “harm widens with study” (insight 1). It also shows that later follow-up can narrow some fears: a faithful use of the case needs both. - BPA is the clearest post-2013 example in the annex of a warning grounded in analogy with an established hazard, the DES mechanism, that later moved a major regulator. - The EU–US split shows that the same evidence can produce opposite regulatory outcomes depending on method choices: intermediate endpoints, and how uncertainty factors are applied.

Claim 8: Silbergeld on antimicrobial growth promoters (pp. 731–733)#

Original claim. - Growth-promoting antimicrobials have “very little support in evidence for efficacy”. The author’s 2007 reanalysis of an industry trial found a “net loss for producers”. - Early national bans did not reduce productivity or harm animal welfare. - “Nearly 80%” of US antimicrobial production goes to food animals. - Use is growing as Brazil, China and India industrialise production. - MRSA was the leading cause of US infectious morbidity and mortality in 2012. - Governments and industry “have little justification to resist change” (pp. 731–733).

Subsequent developments. - Policy moved decisively. - The US FDA completed GFI #213 in January 2017: all production (growth-promotion) indications for medically important antimicrobials were eliminated, and remaining uses were moved to veterinary oversight (FDA timeline, accessed Sep 2026). - China ended growth-promoting antimicrobials in feed from July 2020 (Hu & Cowling, Bull WHO, 2020). - The EU Veterinary Medicines Regulation (EU) 2019/6 (applying from January 2022) restricts prophylactic use. Its Article 118, implemented by Delegated Regulation (EU) 2023/905 and Implementing Regulation (EU) 2024/2598, requires imported animals and products to come from listed countries that do not use antimicrobials for growth promotion, from 3 September 2026. The list includes the US, Brazil and China, among others. - The 2024 UN Political Declaration commits to “meaningfully reduce” antimicrobial use in agri-food systems by 2030 (WHO news, 26 Sep 2024). No numeric agricultural target was adopted. - “Bans don’t hurt productivity”: broadly supported, with nuances. - EU sales of veterinary antimicrobials fell 53% between 2011 and 2022, from 161.2 to 75.8 mg/PCU, in 25 countries reporting throughout (ESVAC final report, Nov 2023). EU livestock production continued meanwhile. - Danish data show productivity rising after the 2000 ban. But total antimicrobial use rose again from 31 to 49 mg/kg of pig by 2008 as therapeutic use increased (Aarestrup et al., AJVR, 2010). Silbergeld’s welfare claim glosses over this. - Efficacy and the “net loss” reanalysis. - Independent economic assessments found small effects rather than a net loss. USDA’s Economic Research Service concluded restrictions were “likely to have small effects on prices and quantities” (Sneeringer et al., ERR-200, 24 Nov 2015). - OECD modelling estimated a global withdrawal would cut meat production by roughly 1% (Laxminarayan, Van Boeckel & Teillant, Feb 2015) [figure not re-verified]. - The “net loss” result has not become a general finding. The “little efficacy” claim holds. - Growth in global use: held. Global veterinary antimicrobial use was estimated at 99,502 t in 2020, projected to rise 8% to 107,472 t by 2030. 67% of hotspots are in Asia (Mulchandani et al., PLOS Glob Public Health, Feb 2023). - Health link. - A systematic review found restricting antibiotic use in food animals associated with 10–15% absolute reductions in resistance in animals. In humans, prevalence of resistance was 24% lower, most clearly among people in direct contact with animals (Tang et al., Lancet Planetary Health, Nov 2017). - Globally, MRSA had the largest increase in AMR deaths from 1990 to 2021: 130,000 attributable deaths in 2021 (GBD AMR Collaborators, Lancet, Sep 2024). - The annex’s claim that MRSA was the single leading US cause of infectious morbidity and mortality in 2012 appears overstated. CDC’s 2019 report puts all AMR infections combined at over 35,000 US deaths a year (CDC AR Threats page, accessed Sep 2026). - The 80% figure. It remains a loosely framed comparison of total kilograms sold across different drug classes (digest caveat). I found no later source that validates it as a like-for-like share.

Verdict: held up. - The policy conclusion that there is “little justification to resist change” has been adopted by the US, the EU and China, and the EU now projects it onto trading partners. - The prediction of rising global use held in aggregate. - The efficacy claim is right in direction (“very little”), but the “net loss” framing is an outlier. - The side claims on animal welfare, the 80% share and MRSA are overstated.

Implications. Strong support for two lessons: “benefit claims need the same scrutiny as risk claims” (insight 5) and “harms that pool in a shared resource make local use a system-wide cost” (insight 9). Post-2013, a further mechanism has become visible and can be stated in general terms: a large market can export its standards through import conditions (Article 118). This counters the displacement dynamic seen in asbestos. The Danish rebound in therapeutic use cautions that banning one use can shift practice to another.

Claim 9: Belpoggi on MTBE (p. 735)#

Original claim. - MTBE is carcinogenic in rats and mice: lymphomas and leukaemias in females, and testicular tumours, in the Ramazzini (CMCRC/RI) model. - MTBE is “probably carcinogenic for humans”. “Children are particularly at risk”. - Precautionary action is warranted before further quantification. - The lab’s earlier formaldehyde findings were later confirmed by IARC (p. 735).

Subsequent developments. - IARC Monograph 138 (evaluation 21 Mar 2025; volume 2026). - MTBE was classified Group 2B, “possibly carcinogenic”, not 2A “probably”. The basis is sufficient evidence in experimental animals, limited mechanistic evidence, and inadequate human evidence (“no informative studies”). - The sufficient animal evidence rests on three GLP inhalation studies: liver tumours in female CD-1 mice, brain astrocytoma in male Wistar rats, and kidney tumours in male F344 rats. - The monograph tabulates the Ramazzini gavage study (Belpoggi et al., 1995). It records increased Leydig-cell (testicular) tumours at the high dose. For the lymphoma/leukaemia finding it notes these “might be mistakenly attributed to treatment-related effects because of significant confounding from respiratory infections” (IARC Vol. 138 MTBE monograph, 2026). - Review of the Ramazzini pathology. An EPA-authored review found close agreement with Ramazzini diagnoses for most tumour types but not for lymphoma/leukaemia involving the respiratory tract, where infections complicated diagnosis (Gift et al., EHP, Sep 2013). - US EPA. Its IRIS programme has still “not assessed” MTBE for carcinogenicity. The last IRIS document dates from October 2000 (IRIS page, accessed Sep 2026). - Related additive. ETBE was also placed in Group 2B, and the related TBA in Group 3 (IARC, Mar 2025).

Verdict: partly held up. Rodent carcinogenicity is now formally accepted, so Belpoggi was right that MTBE is an animal carcinogen. But: - the accepted tumours are largely not the ones she emphasised; - the lymphoma/leukaemia signal from her laboratory is specifically discounted for confounding; - the human classification is “possibly”, not “probably”; - no evidence emerged that children are especially at risk.

Implications. The MTBE case’s durable lesson remains the 2001 one: a groundwater-persistence problem created by a substitute (recorded in the vol. 1 chapter, not this annex). This update is a caution on source reliance. A single-laboratory finding with a disputed diagnostic protocol pointed the right way on animal carcinogenicity but overstated the human case. It should not be cited for any lesson about human cancer risk.

Claim 10: Christoforou on the WTO hormones ruling (pp. 736–737)#

Original claim. This is a “personal summary” by the European Commission’s principal legal adviser of the 16 October 2008 Appellate Body report (US/Canada – Continued Suspension). It calls the report a “significant victory” for the EC. It says the report: - faulted the panel’s expert independence, burden of proof and standard of review; - held that a “restrictive” risk assessment ignored misuse; - held that international standards are not dispositive of “insufficiency” under Article 5.7; - held that a rigid “critical mass” test was wrong.

Final resolution would need agreement or further litigation. The EC’s position is “significantly strengthened”, and the findings will be “highly relevant” to other SPS disputes (pp. 736–737).

Subsequent developments. - Resolved by trade concessions, not litigation. - The US and EU concluded a Memorandum of Understanding on 13 May 2009, revised on 21 October 2013 (notified 14 April 2014). The EU opened a duty-free quota for beef from animals not treated with hormones in exchange for suspension of US retaliatory duties (WTO DS26 page). - In December 2016 the US “took steps to reinstate increased duties” at the request of its beef industry. - The EU then negotiated to allocate a country-specific share of the quota to the US “with a view to a definitive resolution” of DS26 (Commission proposal COM(2019) 297, 2019). - The EU ban was never re-tested in a second round of litigation. - The litigation route closed. The Appellate Body has been unable to hear appeals since December 2019. It has had no members since 30 November 2020 (WTO Appellate Body page, accessed Sep 2026). - Doctrinal influence: confirmed. - The report’s four indicators for reviewing a member’s risk assessment (paras 590–591) were reiterated by the Appellate Body in Australia – Apples (2010). - The report is regularly cited in later SPS panel work such as Costa Rica – Avocados (Mexico). Its correction of the panel’s “restrictive notion of risk assessment” is a standard citation (WTO Analytical Index, SPS Article 5). - The ban persists. The EU’s prohibition on hormone-treated meat remains in force. The EU pays for its maintenance through market access.

Verdict: partly held up. - “Highly relevant to other SPS disputes” held: the standard-of-review test became canonical. - “Significantly strengthened” the EU’s legal position is untested, because the dispute was managed by quota deals and the appellate route has since collapsed. - The “victory” framing, from a party’s own counsel, understated that the Appellate Body also upheld the right of the US and Canada to keep their sanctions in place, as the annex itself concedes (p. 736).

Implications. Supports insight 11 (precaution’s survival in adjudication turns on burden of proof, deference and expert independence) as legal doctrine. The post-2013 history adds a general mechanism: a precautionary measure that trading partners contest may be sustained not by winning the scientific argument but by paying for it with compensatory access. That arrangement is itself vulnerable to renewed pressure, as the 2016 reinstatement move shows. The partisan source means the annex’s account should be read alongside WTO primary texts.

Cross-cutting implications for using this section as a lens#

Stated in technology-neutral terms and tied to the annex’s pages:

  1. Direction versus magnitude (pp. 717–737). The annex’s warnings were more reliable about direction than about size or mechanism: - Direction (vindicated): gasoline, asbestos, BPA, antimicrobial growth promoters, ozone. - Size or mechanism (weak): fish “fully abolish”, PCB epigenetics, MTBE “probable”, benzene 3–9×.

Weight early warnings accordingly: as signals to investigate and to reduce exposure, not as quantitative estimates. 2. Source reliance matters. The claims that failed were mostly those resting on the contributor’s own laboratory, unpublished analyses, or single cohorts (pp. 718, 721–722, 727, 735). The claims that held drew on multi-source assessments (NCI cohorts, WHO, WMO/UNEP, industry trial data). 3. Measurement technology sets the limits (p. 725). EU asbestos law now explicitly requires a move from optical to electron microscopy because the older instrument missed the thinnest harmful fibres. 4. Displacement, and the reverse (pp. 724–726, 731). - Displacement continues: asbestos production concentrated in Russia, Kazakhstan, China and Brazil, with Brazil exporting despite its own constitutional ruling. - A counter-mechanism emerged: a large market extending its production standards to imports (EU Article 118, from September 2026). 5. Legacy and implementation gaps (pp. 724–728). Relief takes decades even after bans: - Asbestos remains in buildings, with EU renovation-linked removal but no deadline. - Over 10 Mt of PCB materials remain, with most countries off track for 2028. - Ozone recovery runs into the 2040s–2060s. - Treaty compliance needs independent detection, as the CFC-11 episode showed. 6. Falling tolerable intakes as the typical form of “expanding harm” (p. 717). The strongest post-2013 evidence for the editors’ thesis is repeated cuts in reference values: dioxins 7×, BPA 20,000×, EU benzene 5×, EU asbestos 10–50×. It is less a matter of wholly new disease endpoints. Where the reference value was contested (BPA), jurisdictions diverged. 7. Precaution in trade law is sustained by bargaining as much as by doctrine (pp. 736–737).

Sources#

All URLs were accessed in September 2026 unless noted. Dates are those of publication or last update.

Carcinogen classification (IARC)#

Benzene and gasoline#

Asbestos#

PCBs and POPs#

Fish, contaminants and nutrition#

Ozone#

DES and BPA#

Antimicrobials#

MTBE#

Hormones and WTO#