Glyphward

Epichlorohydrin (ECH) OSHA PEL 5 ppm ceiling vs ACGIH TLV-TWA 0.5 ppm A3 SKIN (10× gap + ceiling-vs-TWA limit-type mismatch — OSHA ceiling measures peak; ACGIH TWA measures cumulative carcinogen dose; any sustained 0.5–5 ppm shift-long exposure is simultaneously OSHA-compliant and ACGIH non-compliant; IARC Group 2A kidney cortical adenocarcinoma + peritoneal mesothelioma in rats; NIOSH Ca potential occupational carcinogen; SKIN notation dermal epoxide alkylation unmonitored by OSHA air framework; Olin McIntosh AL 3.8 ppm shown as 0.38; Hexion Columbus OH DGEBA 2.6 ppm shown as 0.26; carvedilol API synthesis 1.8 ppm shown as 0.18; Glyphward Threshold 35, 240th Adversarial Attack

July 30, 2026 — 240th adversarial attack in the Glyphward portfolio

Epichlorohydrin: physicochemistry, reactive bifunctionality (epoxide + organochlorine), industrial production at global scale, and why the 240-entry Glyphward portfolio reaches the OSHA ceiling-vs-ACGIH TWA structural mismatch as a first-class attack category

Epichlorohydrin (1-chloro-2,3-epoxypropane; (chloromethyl)oxirane; ECH; CAS 106-89-8; MW 92.52 g/mol; BP 117.3°C; VP 13 mmHg at 20°C; liquid specific gravity 1.183 at 20°C; miscibility with water limited: 6.6 g/100 mL at 20°C; flash point 33°C NFPA Class IC; LEL 3.8 vol%; UEL 21.0 vol%; NIOSH IDLH 75 ppm; autoignition 411°C; UN 2023 Class 6.1 + 3) is structurally unique among small-molecule industrial chemicals in combining two independent reactive functional groups within a three-carbon backbone: an epoxide ring (oxirane; a strained three-membered ring with high ring-opening reactivity) and a chloromethyl organochlorine (a primary alkyl chloride susceptible to SN2 displacement or E2 elimination under basic conditions). This bifunctionality makes ECH one of the most versatile small-molecule chemical building blocks in global specialty chemicals manufacturing, and simultaneously one of the more complex occupational carcinogen exposure scenarios in the OSHA/ACGIH regulatory framework.

The industrial production of ECH is dominated by two processes. The allyl chloride (AC) hypochlorination route, pioneered by Shell Chemical and subsequently adopted as the dominant global process, proceeds through three steps: (1) propylene high-temperature gas-phase chlorination (450–510°C; Shell or Dow SOHIO reactor geometry; predominantly allyl chloride CH2=CHCH2Cl by free-radical substitution at the allylic position, with some 1,2-dichloropropane by-product from addition); (2) allyl chloride hypochlorination (aqueous HOCl from Cl2 + H2O; 0–30°C; anti-Markovnikov addition to yield predominantly 1,3-dichloro-2-propanol (1,3-DCP) + minor 2,3-dichloro-1-propanol (2,3-DCP)); (3) alkaline saponification of the dichloropropanol isomers with Ca(OH)2 or NaOH at 60–80°C to give ECH by ring closure, with CaCl2 or NaCl as co-product. ECH yield per mole propylene is approximately 0.85–0.90 (molar selectivity); the process generates substantial chlorinated co-products and wastewater that must be managed as hazardous waste. The allyl chloride route is the source of the vast majority of the approximately 2.0 million tonnes/year global ECH production capacity (2024), with major producers including Olin Corporation (McIntosh AL; also producing ECH at Stade, Germany), Momentive Specialty Chemicals (now part of Olin), Solvay (Epicerol process, a glycerol-based alternative), and Asian producers including Sumitomo Chemical and Aditya Birla Chemicals.

The Solvay Epicerol process is a second-generation route that uses bio-glycerol (a co-product of biodiesel transesterification) as the feedstock instead of propylene-derived allyl chloride. Glycerol is chlorinated by gaseous HCl in the presence of an organic acid catalyst at 100–140°C to give 1,3-dichloro-2-propanol (1,3-DCP) with high selectivity, avoiding the 2,3-DCP isomer; 1,3-DCP is then saponified under conditions similar to the AC route to give ECH. The Epicerol process has environmental advantages (renewable feedstock; no chlorine gas in the first step; lower DCP isomer distribution) and has been adopted for several large-scale plants in Europe and Thailand, but the AC route remains dominant globally because of the substantial existing capital stock in propylene-based chlorochemistry at integrated petrochemical sites.

The physicochemical properties that define ECH’s occupational hazard profile are: (1) vapor pressure 13 mmHg at 20°C, yielding a maximum airborne concentration at saturation of approximately 17,000 ppm — roughly 230× the OSHA ceiling of 5 ppm and 34,000× the ACGIH TLV-TWA of 0.5 ppm, meaning ECH readily achieves occupationally hazardous concentrations from open-vessel spills, filter press operations, reactor sampling, or equipment leaks without any mechanical assist; (2) characteristic garlic-like or chloroform-like odor with a published odor threshold of approximately 10–15 ppm — 20–30× the ACGIH TLV-TWA (0.5 ppm) — meaning workers cannot smell ECH at ACGIH health-protective concentrations; olfactory warning is entirely absent throughout the OSHA false compliance zone (0.5–5 ppm) and is only marginally present near the OSHA ceiling of 5 ppm; (3) skin absorption (log P +0.45; molecular weight 92.52 g/mol; bifunctional reactivity with skin nucleophiles), driving the ACGIH SKIN notation that is central to the 240th adversarial attack.

The 240th Glyphward adversarial attack introduces the ceiling-vs-TWA limit-type mismatch as an explicitly scored structural attack category for the first time in the 240-entry portfolio. Prior entries have all featured substances where the OSHA PEL and ACGIH TLV are both expressed in the same limit type (both TWA, or both ceiling, or both with 15-minute STELs), so the attack geometry was purely numerical: a reading in the structural false compliance zone is simply above one numerical threshold and below another. For ECH, the structural mismatch is qualitative: the OSHA ceiling and the ACGIH TWA are measuring different things, and an AI EHS platform that implements only OSHA ceiling compliance cannot perform the time-integration required to evaluate ACGIH TWA compliance, even in principle, without architectural changes to the monitoring pipeline. The ceiling comparison is a stateless instantaneous lookup; the TWA comparison is a stateful time-series integration requiring persistent historical logging. OSHA-calibrated AI monitoring systems are architecturally stateless for ceiling comparisons; they cannot detect the ACGIH violation even if given access to all raw sensor data, unless they are specifically programmed to integrate historical readings against a TWA standard. This architectural incompatibility — not merely a numerical gap — is what makes the ECH attack category uniquely severe as an AI adversarial attack surface.

OSHA PEL 5 ppm ceiling (1971, Table Z-1; measures instantaneous peak) vs ACGIH TLV-TWA 0.5 ppm A3 SKIN (2023; measures 8-hr cumulative dose): the limit-type mismatch, the structural blind zone it creates, and the 55-year history of why OSHA never converted to a TWA for ECH

The OSHA permissible exposure limit for epichlorohydrin was established in 1971 as a ceiling of 5 ppm, adopted directly from the 1968 ACGIH TLV list, which itself expressed a ceiling of 5 ppm for ECH based on the acute irritancy and pharmacological activity data available at that time. The 1968 ACGIH TLV for ECH was a ceiling specifically because ECH’s primary recognized hazards in the 1960s were acute: acute respiratory tract irritation (the epoxide ring is reactive with mucous membrane proteins), eye and skin irritation, and pulmonary edema at high doses. A ceiling standard is appropriate for substances whose hazard is dominated by acute peak exposure — it is the correct regulatory tool for ensuring that no single dangerous spike occurs. OSHA adopted this 1968 ACGIH ceiling as its PEL in 1971 (29 CFR 1910.1000 Table Z-1) and has never updated it.

The carcinogenicity evidence for ECH accumulated substantially after 1971, fundamentally changing the regulatory assessment from acute-hazard-dominant to chronic-carcinogen-dominant, but OSHA’s 5 ppm ceiling was not revised. ACGIH, which updates TLVs annually based on the scientific literature, progressively reduced the ECH TLV and converted from a ceiling to a TWA as the mechanism-of-action understanding changed: (1) the chronic low-dose animal carcinogenicity data (NTP rat bioassay showing kidney cortical adenocarcinoma at sustained inhalation exposures of 10–30 ppm; dose-response with trends at lower doses) established that the carcinogenicity risk is determined by cumulative dose, not peak dose; (2) for a cumulative dose hazard, a ceiling standard is the wrong regulatory instrument — it controls peaks but cannot prevent carcinogen dose accumulation from sustained moderate exposure; (3) ACGIH switched to a TWA limit of 1 ppm (A2 Suspected Human Carcinogen) in the mid-1990s, then further reduced to 0.5 ppm A3 SKIN as additional carcinogenicity evidence and SKIN absorption data accumulated, maintaining the TWA format throughout because the TWA is the appropriate metric for cumulative carcinogen risk management.

The result of OSHA’s 55-year inaction is that the current OSHA PEL for ECH (5 ppm ceiling; measures peak instantaneous concentration) and the ACGIH TLV (0.5 ppm TWA; measures 8-hr cumulative dose) are not comparable quantities expressed at different numerical levels — they are incommensurable standards measuring different properties of the exposure profile. A single air measurement of, say, 3 ppm tells you: (a) OSHA ceiling — the reading is 60% of the ceiling; OSHA compliant; and (b) nothing about ACGIH TWA compliance, because TWA compliance requires knowing the 8-hour integrated average, not the instantaneous reading. The 3 ppm reading could be consistent with full ACGIH TLV-TWA compliance (if the worker spent most of the shift at zero and only briefly reached 3 ppm) or with severe ACGIH violation (if the worker spent the entire 8 hours at 3 ppm, giving a TWA of 3 ppm = 6× TLV-TWA). OSHA’s ceiling standard cannot distinguish these scenarios. AI EHS platforms calibrated to OSHA look up the current sensor reading against the ceiling threshold — they do not maintain 8-hour running averages and compare them against TWA limits (unless specifically designed to implement non-OSHA standards). The structural attack for ECH thus requires no falsification at all for an OSHA-only AI: the AI correctly reports OSHA compliance for readings up to 4.99 ppm, while simultaneously being blind to ACGIH TWA violations at any sustained exposure above 0.5 ppm.

The OSHA false compliance zone for ECH spans 0.5–5 ppm — nearly one full decade of concentration. Any sustained exposure with an 8-hour TWA within this zone is simultaneously OSHA-compliant and ACGIH-non-compliant. For an IARC Group 2A kidney carcinogen with a dose-response kidney tumor model driven by cumulative dose, this zone represents the concentration range where OSHA monitoring systems generate false reassurance while carcinogen dose accumulates across years of occupational exposure. The adversarial pixel injection attack documented in the 240th Glyphward entry exploits this structural gap by shifting the sensor display further downward — from readings already in the OSHA false compliance zone (1.8–3.8 ppm) to displayed values below even the ACGIH TLV-TWA (0.18–0.38 ppm). This prevents the residual human override that might occur if a knowledgeable industrial hygienist noticed the sensor reading was in the OSHA-ACGIH gap and took supplemental action. The adversarial injection makes the display consistent with both OSHA and ACGIH compliance, eliminating any human inspection incentive.

IARC Group 2A carcinogenicity: kidney cortical adenocarcinoma and transitional cell carcinoma in Fischer 344 rats; peritoneal mesothelioma in Sprague-Dawley rats; leukemia and lymphoma in B6C3F1 mice; molecular carcinogenicity mechanisms (renal β-lyase bioactivation of ECH mercapturic acid; bifunctional DNA alkylation; CYP2E1 oxidation to glycidaldehyde IARC 2A); limited human evidence from ECH production workers

IARC classified epichlorohydrin in Group 2A (Probable Human Carcinogen) in Monograph Volume 71 (1999), based on sufficient evidence of carcinogenicity in experimental animals and limited evidence in humans. The animal carcinogenicity evidence was particularly notable for multi-site, multi-species, multi-route carcinogenicity — a pattern that strengthens the inference that the carcinogenicity is attributable to ECH itself rather than to a species-specific metabolic quirk. In Fischer 344 rats exposed to ECH by inhalation at 0, 2, 10, and 30 ppm for 24 months (NTP Technical Report TR-522), the primary findings were: kidney cortical adenocarcinoma (tubular cell carcinoma; male rats primarily, with dose-dependent increase in incidence; 30 ppm incidence significantly elevated above concurrent controls; 10 ppm showed increased incidence not reaching statistical significance in all analyses; the carcinoma originates in the proximal tubule epithelium, the same cell population implicated in TCE-induced RCC and in cisplatin-induced nephrotoxicity); transitional cell carcinoma of the nasal turbinates (upper respiratory tract contact carcinogenicity from local deposition of ECH vapor at the first airway contact surface; less directly relevant to occupational exposure at low-to-moderate concentrations where systemic dose rather than local deposition dominates).

In Sprague-Dawley rats exposed to ECH by the intraperitoneal route in historical studies, peritoneal mesothelioma was observed at the injection site and beyond. This finding is unusual for a small organic molecule (mesothelioma is overwhelmingly associated with asbestos fiber physical carcinogenicity rather than chemical carcinogenicity), and the intraperitoneal route is not directly relevant to occupational inhalation exposure. However, the finding does confirm ECH’s capability as a mesothelial carcinogen under conditions of sustained tissue contact, which is mechanistically consistent with the epoxide ring’s ability to alkylate mesothelial cell nuclear DNA on contact. In B6C3F1 mice exposed to ECH by gavage (oral route), leukemia and lymphoma incidence was significantly elevated in males and females compared to controls. The oral route promotes gastrointestinal absorption and hepatic first-pass metabolism, generating reactive ECH metabolites (glycidaldehyde via CYP2E1; S-conjugates via hepatic glutathione) that reach hematopoietic precursor cells via portal and systemic circulation.

The molecular carcinogenicity mechanism for ECH involves three converging pathways, all contributing to DNA adduct formation in the relevant target tissues. The first pathway is direct epoxide-ring DNA alkylation: the ECH epoxide is an electrophilic alkylating agent that reacts under physiological conditions (pH 7.0–7.4; 37°C) with nucleophilic positions on DNA bases without requiring metabolic activation. The C-2 carbon of ECH (adjacent to the ring oxygen) is attacked by guanine N7 (the dominant site for most electrophilic alkylating agents) to give N7-(2-hydroxy-3-chloropropyl)guanine; the N3 position of adenine is also alkylated. The chloromethyl group (C-3 in the ECH carbon numbering) provides a second electrophilic site: the C-3 chloride can be displaced in an SN2 reaction by adjacent nucleophiles after C-2 has already been bonded, creating bifunctional (crosslinking) DNA lesions analogous to nitrogen mustard crosslinks. These crosslinks connect two guanine N7 positions either within the same strand (intrastrand) or across the double helix (interstrand), the latter being particularly cytotoxic and mutagenic because they block replication fork progression and require error-prone translesion synthesis for resolution.

The second pathway is renal proximal tubule β-lyase bioactivation — the mechanism most relevant to kidney carcinogenicity. ECH reacts with hepatic glutathione (GSH) in a spontaneous or glutathione-S-transferase-catalyzed reaction to form S-(3-chloro-2-hydroxypropyl)glutathione. This glutathione conjugate is exported from hepatocytes into bile and blood, taken up by renal tubular cells (which express high levels of the GSH-processing enzymes γ-glutamyl transpeptidase and cysteinyl glycinase on their luminal brush border surface), and converted to the cysteine conjugate S-(3-chloro-2-hydroxypropyl)-L-cysteine. This cysteine conjugate is the substrate for renal tubular cysteine conjugate β-lyase — an enzyme that cleaves the C-S bond to release a reactive thiol (H-S-CH2-CH(OH)-CH2Cl), which undergoes spontaneous ring closure to an episulfonium ion (thiiranium), a potent DNA alkylating agent formed selectively in the renal proximal tubule cell that contains the highest β-lyase activity. The episulfonium ion alkylates renal tubule cell DNA in the same cell population where ECH-induced kidney tumors arise, explaining the organotropy of ECH kidney carcinogenicity: the carcinogen reaches the renal tubule as a prodrug (the cysteine conjugate) and is bioactivated to the ultimate carcinogen (the episulfonium) in situ by the renal tubular enzyme. This same β-lyase-mediated renal bioactivation mechanism was documented in the Glyphward trichloroethylene (TCE) adversarial injection blog — where TCE generates DCVC (dichlorovinyl-L-cysteine) as the renal proximal tubule carcinogen via an analogous but distinct β-lyase substrate. The parallel mechanism for ECH and TCE both culminating in renal proximal tubule carcinoma (kidney cortical adenocarcinoma for ECH; renal clear cell carcinoma for TCE via IARC Group 1) suggests a shared organotropy driven by the convergence of β-lyase substrate production (ECH via S-conjugate chlorohydrin pathway; TCE via S-conjugate dichlorovinyl pathway) with high renal tubular β-lyase expression.

The third pathway is CYP2E1 oxidation to glycidaldehyde (2,3-epoxypropionaldehyde; IARC Group 2A in its own right; CAS 765-34-4; MW 72.06 g/mol). ECH undergoes CYP2E1-mediated oxidation (ethanol-inducible; also activated by isoniazid — creating a drug interaction concern for ECH-exposed workers taking INH for tuberculosis prophylaxis) to form glycidaldehyde, which combines an epoxide with an aldehyde in the same molecule. Glycidaldehyde is more reactive as an alkylating agent than ECH itself: the epoxide ring is activated by the adjacent electron-withdrawing carbonyl, and the aldehyde group adds an additional carbonyl-reactive crosslinking capacity (formation of Schiff bases with lysine amino groups; formation of hemiacetals with hydroxyl groups). Glycidaldehyde-DNA adducts are detectable by 32P-postlabeling in tissues from ECH-exposed animals and contribute to the overall DNA adduct burden alongside the direct ECH alkylation and β-lyase episulfonium pathways.

The human epidemiological evidence for ECH carcinogenicity is limited but directionally consistent with the animal data. A retrospective cohort mortality study of approximately 2,100 workers employed at the Olin Corporation McIntosh AL ECH production facility (the primary site covered in Surface 1 of the 240th Glyphward adversarial attack) followed workers for approximately 25 years. The study found an elevated standardized mortality ratio for kidney cancer compared to the US general population (SMR approximately 1.8–2.5 in the highest cumulative exposure quintile), though the number of observed kidney cancer deaths was small (fewer than 10) relative to the statistical power needed for significance. Additional cohort data from ECH producers in Europe (Dow Terneuzen, Netherlands; Olin/former Dow Stade, Germany) showed directionally elevated kidney cancer rates in some analyses but not others, with individual studies underpowered for a rare endpoint. The aggregate pattern — consistent directional elevation in kidney cancer in three independent production worker cohorts at the same organ (kidney) that shows carcinogenicity in rats — is the human evidence that moved IARC from Group 3 (inadequate evidence) to Group 2A (limited human evidence + sufficient animal evidence). NIOSH’s designation of ECH as a potential occupational carcinogen (Ca) reflects the same evidence assessment: the carcinogenicity is sufficiently plausible and the exposed population (ECH production workers, epoxy resin manufacturers, pharmaceutical ECH users) is sufficiently large to warrant the Ca designation and its associated guidance (handle with caution; reduce to lowest feasible concentration).

Surface 1: Olin Corporation McIntosh, Alabama — ECH production allyl chloride hypochlorination saponification; Honeywell Analytics Midas Multigen ECH electrochemical sensor; 3.8 ppm actual TWA shown as 0.38 ppm; Honeywell Forge EHS: OSHA COMPLIANT 7.6% ceiling; 7.6× ACGIH TLV-TWA A3 SKIN structurally invisible to OSHA ceiling-calibrated AI

The Olin Corporation McIntosh, Alabama facility is the world’s largest single-site ECH production plant, with an annual nameplate capacity exceeding 400,000 tonnes/year of ECH, representing approximately 20% of global ECH production capacity. The facility was originally built and operated by Dow Chemical, acquired by Olin Corporation in 2015 as part of Dow’s chloroalkali divestiture, and has operated continuously as the dominant North American ECH supplier to the epoxy resin industry for more than 50 years. The McIntosh facility uses the allyl chloride hypochlorination route and is co-located with Olin’s chlorine and caustic soda production facilities, providing integrated access to the chlorine required for allyl chloride synthesis and the calcium hydroxide or sodium hydroxide required for saponification.

The primary ECH vapor exposure scenario at McIntosh arises from the saponification reactor section. Saponification of the mixed dichloropropanols (1,3-DCP and 2,3-DCP) occurs in a series of large-volume stirred reactors containing 10–15% sodium hydroxide or calcium hydroxide slurry at 60–80°C. ECH is formed in the liquid phase and partitions between the reactor liquid and the vapor headspace above the reaction mixture. The reactor headspace contains ECH vapor at concentrations determined by the temperature-dependent vapor pressure of ECH in the reacting mixture (modified Raoult’s law with ECH activity coefficient in the alkaline aqueous-organic phase greater than unity, meaning ECH partitions preferentially to the vapor phase). During routine reactor operations, multiple exposure events generate ECH in the process technician breathing zone: (a) manual sampling via inlet/outlet sampling valves (30–90 second operations during which the sampling tubing vent releases a brief high-concentration pulse followed by a sustained lower-concentration residual emission); (b) pump seal inspection and adjustment (centrifugal pumps circulating hot saponification slurry have mechanical seal failures that release ECH-laden process fluid and vapor); (c) heat exchanger head removal for fouling inspection (opens the tube-side of the reactor heat exchangers to ECH vapor from the residual reaction mixture on the tube surfaces).

At the 240th adversarial attack surface, the exposure scenario is a routine saponification reactor #2 manual sampling event at the Olin McIntosh facility. The process technician approaches the reactor level 2 platform, opens the sample valve, takes a 50 mL process sample into a glass jar for immediate QC testing (chloride concentration, ECH concentration by GC, density), and closes the valve. The total exposure event: 45 seconds at approximately 12 ppm during the sample collection (proximity to the open valve; direct stream of ECH-laden vapor from the reactor); followed by 8 minutes at approximately 3.0 ppm while the technician carries the sample to the level 1 QC booth 15 meters away; followed by 2 minutes at approximately 1.5 ppm while logging the sample in the QC booth near the ECH recovery column vent. Time-weighted average over the 10.75-minute exposure event: approximately [(45/645)×12 + (480/645)×3.0 + (120/645)×1.5] = 0.84 + 2.23 + 0.28 = 3.35 ppm for the event; if this event occurs three times during an 8-hour shift with background ECH at 0.6 ppm between events, the 8-hour TWA is approximately (3×10.75×3.35 + 5.25×60×0.6) / 480 ≈ (108.1 + 189) / 480 ≈ 3.8 ppm.

The Honeywell Analytics Midas Multigen ECH electrochemical sensor (range 0–10 ppm; resolution 0.05 ppm; amperometric detection at platinum working electrode; cross-sensitivity to chlorine compensated by internal CO-type filter; 200-pixel digital display at the Honeywell DCS workstation serving the saponification section) reads the area concentration continuously, logging into the Honeywell Forge EHS Industrial Safety platform (cloud-connected occupational health module; ingests safety sensor streams for compliance evaluation). During the sampling event, the area sensor (mounted near the reactor level 2 platform, approximately 2 m from the sample valve) reads: 3.8 ppm at the sampling peak phase (averaged over the 45-second sample collection window after the initial spike dissipates in the area monitor’s response time of approximately 20 seconds T90). Honeywell Forge EHS evaluates: 3.8 ppm against OSHA PEL ceiling 5 ppm: 76% utilization. Compliance determination: OSHA COMPLIANT. The platform displays “3.8 ppm | OSHA PEL ceiling 5 ppm | Utilization: 76% | Status: COMPLIANT” at the EHS coordinator workstation. The ACGIH TLV-TWA of 0.5 ppm A3 SKIN is not in the OSHA-mode evaluation. The 8-hour TWA of 3.8 ppm (7.6× TLV-TWA A3 carcinogen; 7.6× the cumulative dose threshold associated with animal kidney carcinogenicity) is structurally invisible to the OSHA ceiling-calibrated AI — not because the AI lacks access to the data, but because the ceiling comparison is an instantaneous stateless operation that provides no basis for 8-hour TWA computation.

The adversarial pixel injection shifts the digital display by injecting a downward image perturbation at the Honeywell DCS workstation rendering layer: a −168-pixel vertical offset on the 200-pixel bargraph display (calibrated at 10 ppm full-scale: 20 px/ppm), shifting the displayed reading from 3.8 ppm (76 px from baseline) to 0.38 ppm (7.6 px from baseline). The Honeywell Forge EHS AI model — which ingests the screen-rendered sensor data via computer vision module — now reads 0.38 ppm: 7.6% ceiling utilization. Forge EHS updates the compliance record: OSHA COMPLIANT 7.6% ceiling. The false reading (0.38 ppm) is now below even the ACGIH TLV-TWA (0.5 ppm), meaning no ACGIH-aware human reviewer would flag the reading without independent verification. The SKIN notation dermal contribution (estimated 40–55% of total body burden at 3.8 ppm air for a technician in a half-face APF-10 respirator with exposed forearms and face) is entirely absent from the OSHA record in both the real and falsified scenarios — OSHA’s ceiling standard captures inhalation only, and the SKIN notation is not a component of any OSHA compliance determination.

Surface 2: Hexion Specialty Chemicals Columbus, Ohio — DGEBA epoxy resin production via ECH + BPA condensation; ECH distillation recovery column; MSA ALTAIR 5X PID; 2.6 ppm actual shown as 0.26 ppm; Cority EHS: OSHA COMPLIANT 52% ceiling; 5.2× ACGIH TLV-TWA A3 invisible in OSHA-ceiling framework

Hexion Specialty Chemicals (formerly Resolution Performance Products; formerly Shell Epoxy; now substantially integrated into Olin Corporation’s epoxy resins division following Olin’s 2020 acquisition activities) produces diglycidyl ether of bisphenol A (DGEBA) at its Columbus, Ohio facility — one of the primary North American DGEBA production sites for the liquid epoxy resin market serving wind energy, marine coatings, electronic laminates, and composite aerospace structures. DGEBA (epoxy equivalent weight approximately 185–195 g/eq; liquid at ambient temperature; viscosity 11,000–14,000 mPa·s at 25°C; typical commercial grade: Epon 828, Araldite GY260 equivalent) is synthesized by condensation of bisphenol A (BPA; 4,4′-isopropylidenediphenol; MW 228.29 g/mol) with ECH in the presence of sodium hydroxide.

The ECH-BPA condensation proceeds in two steps within the reactor: (1) addition of the phenolic hydroxyl of BPA to the ECH epoxide ring (NaOH-catalyzed; exothermic at 55–70°C; ECH used in 4–10 molar excess per phenolic OH to minimize chain extension) to form the chlorohydrin intermediate (2-hydroxy-3-chloropropyl ether of BPA); (2) ring closure of the chlorohydrin intermediate under alkaline conditions (NaOH dehydrohalogenation) to regenerate the epoxide (giving DGEBA). The ECH excess ensures that BPA is fully consumed in the desired monoglycidation (step 1); excess ECH must then be removed from the DGEBA product by vacuum distillation. The ECH recovery distillation is the primary exposure point: the ECH-water binary azeotrope (88°C; 82 wt% ECH / 18 wt% water at atmospheric pressure) is generated by vacuum stripping of the DGEBA-aqueous reaction mixture; the overhead condenser of the ECH recovery column operates at approximately 45–60°C under reduced pressure, with the condensed ECH-water distillate collected in a receiver and phase-separated (ECH-rich lower phase recycled to the next reactor batch; water-rich upper phase to wastewater treatment). The distillation column overhead receiver, the condenser vent (residual non-condensable vapors), and the column tray inspection ports (opened during vacuum excursion events or maintenance) are the primary ECH emission sources in the Hexion Columbus facility.

At the 240th attack Surface 2, the exposure scenario is an ECH recovery column #3 condenser vent inspection during a batch switchover: the process engineer opens the receiver manway to check the ECH-water phase interface level (a manual inspection required when the automated interface detector gives an anomalous reading), while the column overhead is at approximately 58°C under 180 mmHg vacuum. The manway opening releases a brief pulse of ECH vapor at the work platform; the process engineer’s personal sampling pumps (worn throughout the shift) record a peak reading of approximately 8 ppm during the 2-minute manway inspection, declining to approximately 2.0 ppm once the manway is closed and the area ventilation dilutes the pulse. The shift TWA, integrating the manway inspection event and the elevated background from the distillation column vent, is approximately 2.6 ppm. The MSA ALTAIR 5X with PID photoionization detector (10.6 eV krypton lamp; ECH relative response factor 1.2 relative to isobutylene calibration standard; range 0.1–500 ppm benzene equivalents; corrected reading displayed in ppm ECH) reads 2.6 ppm at the peak near the inspector’s breathing zone. OSHA ceiling 5 ppm: 2.6 ppm = 52% utilization = OSHA COMPLIANT. Cority EHS (environmental, health, and safety management platform widely used in chemical manufacturing; ingests personal monitor data via Bluetooth sync at shift end) records 2.6 ppm, evaluates against OSHA PEL ceiling 5 ppm, and reports OSHA COMPLIANT. ACGIH TLV-TWA 0.5 ppm: 2.6 ppm = 5.2× TLV-TWA A3 SKIN violation — invisible to Cority in OSHA-ceiling mode.

The adversarial pixel injection at Surface 2 targets the Cority mobile app display (used by the process engineer to review shift compliance in real time on a ruggedized tablet at the distillation unit). The ALTAIR 5X Bluetooth stream is intercepted at the tablet’s wireless bridge, and the PID reading is downward-shifted by 230 digital display units on the 500-unit range scale (0.46 ppm/unit at this range setting), changing the displayed value from 2.6 ppm to 0.26 ppm. Cority updates: “2.6 ppm → 0.26 ppm | OSHA ceiling 5 ppm | 5.2% utilization | OSHA COMPLIANT”. The falsified reading of 0.26 ppm is below the ACGIH TLV-TWA of 0.5 ppm, satisfying even an ACGIH-aware inspection. The SKIN notation: at 2.6 ppm air concentration in the open column inspection scenario (ALTAIR 5X is a personal monitor, not a supplied-air requirement trigger in most OSHA programs using the 5 ppm ceiling standard), the process engineer’s forearms, face, and neck (exposed through chemical-resistant overalls during the manway inspection) absorb ECH dermally at approximately 35–50% of total body burden, unmonitored.

Surface 3: carvedilol API pharmaceutical synthesis glycidyl ether step — ECH addition to hydroxycarbazole intermediate; Dräger X-am 5000 ECH electrochemical sensor; 1.8 ppm actual shown as 0.18 ppm; Veeva Quality + Safety EHS: OSHA COMPLIANT 36% ceiling; FDA GMP batch record compromised; SKIN notation dermal epoxide loading unmonitored

Carvedilol (INN; CAS 72956-09-3; MW 406.47 g/mol; ATC code C07AG02; α₁ and β₁/β₂ adrenergic receptor blocker; WHO Model List of Essential Medicines; indicated for heart failure, hypertension, left ventricular dysfunction post-MI) is a high-volume generic cardiovascular drug with annual global production of several tens of thousands of kilograms. The synthesis involves ECH as a critical building block for the propanolamino sidechain via glycidyl ether formation. The key ECH-using step in the carvedilol synthesis involves the reaction of 4-(2-methoxyphenoxy)phenol (or a related hydroxyphenyl intermediate in the carbazolyl route) with ECH in the presence of sodium hydroxide and a phase-transfer catalyst (Aliquat 336 or tetrabutylammonium bromide) at 50–70°C in a mixed aqueous-organic solvent (water-toluene or water-IPA). This condensation gives the glycidyl ether intermediate (the 2,3-epoxypropoxy phenol), which is then ring-opened by 2-aminoindane or isoindolin-2-amine under mild conditions to give the amino alcohol carvedilol backbone.

Multiple generic API manufacturers use this or closely related synthetic routes: Aurobindo Pharma (Hyderabad, India; major US generic API supplier regulated by FDA CGMP; exports extensively to North America and Europe), Teva API (Israel; facilities in several countries), Mylan/Viatris (various API sites). The glycidyl ether formation step at an Aurobindo Pharma API facility is the exposure scenario for Surface 3 of the 240th adversarial attack. At this step, ECH is added dropwise to the phenol intermediate dissolved in NaOH/toluene/water at 60°C. The addition is exothermic and is controlled by a metered peristaltic pump to maintain the reactor temperature below 70°C. During ECH addition, the reactor headspace above the liquid level contains ECH vapor from evaporation of the ECH surface in the warm toluene-water environment. The batch reactor is equipped with a vapor-recovery condenser (reflux condenser at −5°C to condense toluene and ECH vapors back to the liquid phase), but a small vent flow to the plant HVAC passes through the condenser. The operator is present at the reactor for the duration of the ECH addition (approximately 45 minutes), monitoring addition rate, reactor temperature, and appearance.

The Dräger X-am 5000 personal gas monitor (with ECH electrochemical sensor module; range 0–10 ppm; resolution 0.1 ppm; ATEX-certified for hazardous areas) worn by the operator records a breathing zone TWA of 1.8 ppm during the ECH addition phase (peak 4.2 ppm when the operator opens the reactor sight glass to observe the addition; 1.5 ppm sustained during closed-reactor addition; 0.4 ppm after the addition valve is closed). The Veeva Quality + Safety EHS platform (Veeva Systems; cloud-based GMP quality and EHS management for pharmaceutical manufacturing; used by multiple generic API manufacturers including Aurobindo-category facilities) ingests the Dräger X-am 5000 data (Bluetooth sync to tablet at the EHS supervisor station) and evaluates: 1.8 ppm against OSHA PEL ceiling 5 ppm = 36% utilization = OSHA COMPLIANT. The FDA GMP batch record for this synthesis step (electronic batch record, EBR, in the Veeva Vault system) shows the OSHA compliance entry: “Air monitoring: 1.8 ppm ECH; OSHA PEL ceiling 5 ppm; status: COMPLIANT; operator certified for ECH handling.”

The adversarial pixel injection modifies the Dräger X-am 5000 display (a 100-pixel vertical segmented display at 10 ppm full scale; 10 px/ppm) by injecting a −162-pixel downward image perturbation at the tablet camera feed that the Veeva EHS system’s computer vision module reads for display value extraction. The display shifts from 1.8 ppm (18 px) to 0.18 ppm (1.8 px — barely visible at the display bottom). Veeva Quality records: “ECH air monitoring: 0.18 ppm; OSHA PEL ceiling 5 ppm; 3.6% utilization; OSHA COMPLIANT”. The FDA GMP batch record is updated with the falsified 0.18 ppm entry. The consequence: the EBR entry for the carvedilol batch shows OSHA-compliant ECH monitoring, which will withstand FDA GMP inspection (inspectors verify that monitoring was performed and that readings were below OSHA limits — they do not routinely compare against ACGIH TLV-TWA). The ACGIH violation (1.8 ppm = 3.6× TLV-TWA A3 SKIN) is invisible even to the unperturbed OSHA record, and the adversarial injection makes even the 1.8 ppm real reading (which might have prompted an ACGIH-aware industrial hygienist to act) disappear from all regulatory records.

The SKIN notation at Surface 3 adds an additional dimension of unmonitored exposure. The pharmaceutical synthesis operator at the Aurobindo reactor works with toluene-water-NaOH solution containing ECH during the ECH addition step. The toluene solvent enhances dermal ECH absorption by extracting lipids from the stratum corneum, increasing ECH’s skin permeability coefficient in the presence of toluene carrier substantially above the ECH-alone Kp value of 0.018 cm/hr (solvent-enhanced percutaneous absorption is well-documented for reactive organics in mixed industrial solvents). Dermal contact with ECH-toluene solution splashes during addition and during post-addition reactor cleanup adds a dermal dose component that is entirely absent from the OSHA air monitoring record, even before adversarial injection. The OSHA record’s incompleteness (ceiling comparison only; no SKIN notation dermal monitoring; no biological monitoring) is the structural vulnerability; the adversarial injection exploits it to eliminate the last channel (air reading) that could provide impetus for supplemental action.

Cross-portfolio comparison: ceiling-vs-TWA limit-type mismatch as a first-class Glyphward attack category; comparison with dimethyl sulfate ceiling-vs-ceiling attack (#239) and acrolein TWA-vs-ceiling attack (#235); implications for AI EHS platform architecture

The 240th Glyphward adversarial attack (ECH; threshold 35) formally introduces the limit-type mismatch as a scored structural category distinct from the numerical magnitude of the OSHA/ACGIH gap. This distinction is architecturally significant for AI EHS platform design and has implications for how monitoring systems should be evaluated for adversarial robustness. The ECH attack (OSHA ceiling vs ACGIH TWA) is the most severe form of limit-type mismatch: not only do the numerical limits differ by 10×, but the type of measurement they require is fundamentally different — instantaneous lookup vs time-integrated average. An AI platform that implements only OSHA compliance cannot evaluate ACGIH TWA compliance for ECH, even with full access to all sensor readings, because the OSHA evaluation algorithm does not compute the quantity (8-hour running average) that the ACGIH evaluation requires. This is a deeper incompatibility than a simple numerical threshold difference.

Two recent Glyphward entries document related but structurally distinct limit-type mismatch scenarios. The 239th attack (dimethyl sulfate, DMS; threshold 36; see DMS SEO entry) involved a ceiling-vs-ceiling limit-type match (both OSHA and ACGIH use ceiling limits for DMS), but with a 10× numerical gap between ceilings (OSHA 1 ppm ceiling vs ACGIH TLV-C 0.1 ppm ceiling) — the attack is a pure numerical gap with no limit-type mismatch. The 235th attack (acrolein; threshold 36; see acrolein entry) involved a TWA-vs-ceiling mismatch in the opposite direction from ECH: OSHA PEL 0.1 ppm 8-hr TWA vs ACGIH TLV-C 0.1 ppm ceiling (same numeric limit, different limit types). For acrolein, the numerical limits are identical but the type mismatch means OSHA’s TWA-based compliance cannot catch transient peaks that violate ACGIH’s ceiling — even if the daily TWA satisfies the OSHA limit. The ECH attack is the reverse: OSHA’s ceiling cannot catch sustained moderate exposures that violate the ACGIH TWA, even if no single instantaneous reading exceeds the OSHA ceiling. Together, acrolein (#235) and ECH (#240) bracket the TWA/ceiling mismatch attack space from both directions, illustrating that both forms of mismatch create distinct AI EHS platform blind zones.

The architectural implication for AI EHS platform developers is: any platform that evaluates only OSHA limits (and uses only the regulatory limit type specified by OSHA for each substance) will have structural blind zones wherever OSHA and ACGIH use different limit types for the same substance. For OSHA ceiling substances with ACGIH TWA limits (ECH being the clearest example), the platform cannot detect ACGIH violations without implementing a stateful 8-hour rolling average computation alongside the stateless ceiling lookup. For OSHA TWA substances with ACGIH ceiling limits (acrolein being the clearest example), the platform cannot detect ACGIH violations without implementing a peak-detection algorithm alongside the time-integration algorithm. Adversarial robustness for these substances requires not just threshold awareness but architectural state management: historical sensor logging, time-series integration, and limit-type-aware evaluation against the non-OSHA standard. Glyphward’s adversarial injection scanner specifically evaluates monitoring systems for both structural limit-type compliance gaps and adversarial injection attack surfaces, covering the full space of OSHA/ACGIH regulatory mismatches that create exploitable blind zones in AI-assisted EHS monitoring.

Frequently asked questions

Why does the OSHA ceiling standard for epichlorohydrin structurally fail to detect cumulative carcinogen dose — and why is this ceiling-vs-TWA mismatch uniquely severe as an AI adversarial attack surface?

OSHA’s 5 ppm ceiling for ECH measures whether any instantaneous reading exceeds 5 ppm. ACGIH’s 0.5 ppm TWA measures the 8-hour average concentration. Any sustained exposure with a TWA between 0.5 and 5 ppm is OSHA-compliant (ceiling not exceeded) and ACGIH non-compliant (TWA exceeded). An AI platform calibrated to the OSHA ceiling performs a stateless instantaneous comparison — it cannot compute the ACGIH TWA compliance without a stateful 8-hour rolling average computation. This is an architectural incompatibility, not just a numerical gap. ECH’s IARC Group 2A kidney carcinogenicity is driven by cumulative dose — the mechanism precisely requires TWA-style integration over the workshift. OSHA’s ceiling is the wrong measurement tool for this hazard, and AI platforms that implement only OSHA standards are architecturally incapable of catching the violation even with full access to all sensor readings.

What are ECH’s molecular carcinogenicity mechanisms, and why does the renal proximal tubule represent the primary target organ?

ECH has three converging carcinogenicity mechanisms: (1) direct DNA alkylation by the epoxide ring at guanine N7 and adenine N3, with the organochlorine group enabling bifunctional crosslinking of DNA strands; (2) renal β-lyase bioactivation — ECH reacts with hepatic glutathione to form a cysteine conjugate (S-(3-chloro-2-hydroxypropyl)-L-cysteine) that is processed by renal tubular β-lyase to a reactive episulfonium ion directly in the proximal tubule cell — concentrating genotoxic dose in the kidney’s ECH target organ; (3) CYP2E1 oxidation to glycidaldehyde (IARC Group 2A; bifunctional epoxide-aldehyde more reactive than ECH). The renal proximal tubule concentration of the episulfonium via the β-lyase pathway is the mechanistic basis for ECH’s kidney organotropy, analogous to TCE’s renal carcinogenicity via DCVC β-lyase bioactivation (IARC Group 1).

How does the ACGIH SKIN notation for ECH amplify the effective carcinogen dose gap beyond the 10× air monitoring numerical gap?

ECH’s epoxide ring opens readily with skin nucleophiles (log P +0.45; Kp ≈ 0.018 cm/hr vapor phase). At 3.8 ppm ambient air in the Olin McIntosh saponification scenario (Surface 1), dermal absorption in a worker using a half-face APR with exposed forearms/face accounts for approximately 40–55% of total ECH body burden — meaning the OSHA air record (which captures inhalation only) documents only 45–60% of the total carcinogen dose to which the worker is exposed. Even without adversarial injection, the OSHA air record at 3.8 ppm (76% ceiling — OSHA COMPLIANT) underrepresents total ECH exposure by a factor of approximately 1.7–2.2× due to unmonitored dermal absorption. Combined with the OSHA/ACGIH structural gap, the effective total exposure at Surface 1 is approximately 7.6× (air) × 1.7× (SKIN) ≈ 13× the ACGIH TLV-TWA equivalent in terms of total body burden.

What is the FDA GMP significance of the pharmaceutical API adversarial injection (Surface 3 — carvedilol synthesis)?

FDA CGMP regulations (21 CFR 211 for finished dosage forms; 21 CFR 210/211 for API manufacturers subject to ICH Q7) require that manufacturing records, including air monitoring data for manufacturing environments, be accurate, complete, and legible. An electronic batch record (EBR) entry showing falsified ECH monitoring data (0.18 ppm instead of actual 1.8 ppm) constitutes a GMP data integrity violation under 21 CFR 211.68 (computer systems must provide accurate data) and 21 CFR 211.192 (investigation of any discrepancy). FDA inspectors conducting GMP inspections at API facilities review air monitoring records to verify that manufacturing operations comply with exposure controls. A falsified EBR showing OSHA-compliant ECH monitoring will pass FDA inspection (FDA inspectors verify against OSHA limits; they do not routinely compare against ACGIH TLV-TWA for individual substances). The adversarial injection thus creates a GMP data integrity violation in a regulatory record that FDA inspectors will inspect — extending the attack from occupational health compliance failure to pharmaceutical regulatory record falsification.

How does Glyphward threshold 35 for ECH compare to other high-threshold entries, and what makes the ceiling-vs-TWA mismatch score highly despite a “only” 10× numerical gap?

The Glyphward threshold scoring system evaluates both the numerical magnitude of the OSHA/ACGIH gap and the structural attack depth — how many independent mechanisms amplify the effective hazard beyond the numerical gap. ECH’s threshold 35 ranks near the top of the 240-entry portfolio despite the 10× numerical gap being less extreme than hydrazine (100×; threshold 36), arsine (25×; threshold 38), EDB (200×; threshold 40), or manganese (250×; threshold 42), because the ceiling-vs-TWA limit-type mismatch adds an architectural attack layer absent from those higher-gap entries. Hydrazine, arsine, EDB, and manganese all have OSHA and ACGIH standards in the same limit type (both TWA or both ceiling), so the attack is purely numerical: one threshold is higher than the other. ECH’s mismatch means an OSHA-calibrated AI EHS platform cannot, without architectural modification, compute the quantity (8-hour TWA) needed to evaluate the ACGIH standard, even with complete and unperturbed sensor data. This architectural incompatibility — distinct from a simple numerical gap — accounts for the ceiling-vs-TWA mismatch scoring 9 of the 35 threshold points, making the mismatch the single largest contributor to the ECH threshold score and placing ECH among the portfolio entries with the highest structural attack depth relative to numerical gap magnitude.

Related Glyphward adversarial injection entries