Adversarial Injection · Vinyl Chloride Monomer (VCM) PVC Polymerization, Transfer Operations, and Hepatic Angiosarcoma Ultrasound AI Monitoring · Attack #196

Vinyl Chloride Monomer (VCM; Chloroethylene; CAS 75-01-4; MW 62.50 g/mol; BP −13.4°C; LEL 3.6%; Vapor Pressure 2,890 mmHg at 20°C; Odor Threshold 500–5,000 ppm — CANNOT Detect at PEL by Smell) PVC Polymerization, Autoclave Confined Space Entry, and VCM Transfer Operations at Westlake Chemical, Formosa Plastics USA, OxyChem/Olin, and Shintech — OSHA 29 CFR 1910.1017 PEL 1 ppm TWA / Ceiling 5 ppm / Action Level 0.5 ppm (OSHA's FIRST Substance-Specific Carcinogen Standard; Enacted 1974 — B.F. Goodrich Louisville KY Four PVC Kettle Cleaner Angiosarcoma Deaths), ACGIH TLV-TWA 1 ppm A1 (Same Numerical Limit as OSHA PEL — Both Organizations Agree; Unique Structural Vulnerability: OSHA 15-Min Ceiling 5 ppm vs. ACGIH A1 Notation Treating Any >1 ppm Exposure as Confirmed Carcinogen), NIOSH Ca REL Lowest Feasible Concentration, IARC Group 1 Hepatic Angiosarcoma of the Liver (ASL; 400–500× Background Rate in VCM Cohort; Median Survival Advanced ASL <6 Months; 100% CFR Without Surgical Resection; 20–40 Year Latency; CYP2E1 → Chloroethylene Oxide → εA/εC DNA Adducts → p53 Codon 179/249 G→A VCM Mutation Fingerprint), OSHA 1910.1017(k) Medical Surveillance (Hepatic Function Monitoring; Annual LFTs/Ultrasound), Confined Space Autoclave Pre-Entry VCM Air Monitoring, and Hepatic Angiosarcoma Ultrasound AI Diagnostic Falsification: AI Prompt Injection via Pixel Perturbation — FIRST VCM AI Attack

Vinyl chloride monomer (VCM; chloroethylene; CAS 75-01-4; MW 62.50 g/mol; colorless flammable gas; BP −13.4°C; LEL 3.6% UEL 33%; vapor pressure 2,890 mmHg at 20°C; faintly sweet ether-like odor but odor threshold 500–5,000 ppm — workers cannot reliably detect VCM at or near the OSHA PEL 1 ppm by smell) is the primary monomer for polyvinyl chloride (PVC) production, one of the highest-volume synthetic materials globally (>40 million tons PVC/year). OSHA 29 CFR 1910.1017 PEL: 1 ppm TWA; ceiling 5 ppm (any 15-min period); action level 0.5 ppm TWA. OSHA 1910.1017 is OSHA's FIRST substance-specific carcinogen standard, enacted in 1974 after four PVC autoclave kettle cleaners at B.F. Goodrich Louisville KY were diagnosed with hepatic angiosarcoma of the liver (ASL) — three died within twelve months. ACGIH TLV-TWA: 1 ppm A1 (same numerical limit as OSHA PEL; both organizations agree on 1 ppm — unlike asbestos where OSHA is 10× more protective; the unique structural vulnerability is the OSHA 15-min ceiling allowance of 5 ppm vs. ACGIH's A1 notation treating any exceedance above 1 ppm as an exposure to a confirmed human carcinogen). NIOSH Ca REL: lowest feasible concentration. IARC Group 1: hepatic angiosarcoma (ASL; 400–500× background rate in VCM-exposed cohorts; 1–2 cases/million/year background; 20–40 year latency; median survival from advanced ASL diagnosis under 6 months; 100% case-fatality rate without surgical resection of early lesions). Critical mechanistic pathway: VCM → CYP2E1 hepatic oxidation → chloroethylene oxide (CEO; reactive epoxide) and chloroacetaldehyde (CAA; CEO rearrangement) → cyclic DNA adducts (N2,3-ethenoguanine εG; 1,N6-ethenoadenine εA; 3,N4-ethenocytosine εC) → G→A transitions at p53 codons 179 and 249 — the VCM-specific mutation fingerprint distinguishable from spontaneous p53 mutations by molecular epidemiology.

The VCM regulatory framework has a structural uniqueness within the Glyphward portfolio: OSHA PEL 1 ppm and ACGIH TLV-TWA 1 ppm are numerically identical — unlike asbestos (OSHA 10× more protective at 0.1 f/cc vs. ACGIH 1 f/cc TLV-TWA) or 1,3-butadiene (ACGIH 2 ppm more protective than OSHA 1 ppm). Both regulatory bodies converge at 1 ppm TWA for VCM. The unique structural vulnerability for adversarial AI exploitation is therefore not a concentration-limit gap between agencies but rather two other regulatory asymmetries: first, the OSHA 15-min ceiling (5 ppm) creates a 5× window above the TWA PEL where brief peak exposures are OSHA-permissible but ACGIH's A1 confirmed carcinogen notation implies no safe level above 1 ppm; second, the OSHA 1910.1017 action level (0.5 ppm) triggers mandatory medical surveillance and enhanced monitoring requirements — an adversarial AI that falsifies personal monitoring below the 0.5 ppm action level (even while actual concentrations exceed the 1 ppm PEL) simultaneously bypasses both the PEL enforcement obligation and the medical surveillance activation threshold that OSHA specifically enacted as the clinical detection mechanism for early VCM-associated angiosarcoma. The B.F. Goodrich historical pathway — confined space autoclave kettle cleaners with high VCM exposure accumulated over years → ASL latency 20–40 years — is directly replicated by Surface 1 of this attack: AI falsification of TVA-2020 FID/PID pre-entry monitoring at a Westlake Chemical PVC polymerization autoclave authorizes confined space entry with an air-purifying respirator when actual VCM 4.8 ppm (at ceiling 5 ppm) requires supplied-air or SCBA under OSHA 1910.1017(j)(2).

Why PVC Polymerization, VCM Transfer Operations, and VCM-Cohort Hepatic Surveillance Are Disproportionately Vulnerable to AI Monitoring Attacks

Vinyl chloride monomer operations and VCM-cohort medical surveillance have five structural vulnerabilities that amplify adversarial AI monitoring attacks beyond those of most other regulated substances. First, OSHA 1910.1017(d) requires continuous air monitoring in all areas where VCM may be present — the monitoring infrastructure is mandatory and comprehensive, meaning adversarial AI integration into mandatory monitoring platforms directly undermines the regulatory baseline rather than only voluntary supplemental monitoring. The VCM monitoring requirement is more comprehensive than most substance-specific OSHA standards precisely because OSHA learned from the B.F. Goodrich failure: the kettle cleaners accumulated high VCM doses over years without any quantitative exposure monitoring. Second, the PVC polymer matrix itself creates an insidious exposure hazard: VCM monomer is physically trapped within PVC latex during batch polymerization and does not release fully during standard N₂ purge cycles; autoclave pre-entry VCM residuals depend on batch conversion percentage, solids content, and purge duration in ways that are difficult to predict from external engineering calculations alone. An autoclave that appears to have completed a 3-cycle nitrogen purge may still contain 4–8 ppm VCM liberated by PVC wall scraping during maintenance — exactly the scenario Surface 1 describes. Third, VCM has no reliable olfactory warning at or near the PEL: the odor threshold ranges from 500 to 5,000 ppm across individuals, meaning the OSHA PEL of 1 ppm is 500–5,000× below the minimum detectable concentration by smell. Workers entering a VCM-regulated area have no sensory backup — air monitoring is the only real-time warning, and adversarial AI falsification of that monitoring eliminates the sole warning pathway entirely. Fourth, the 20–40 year latency between first VCM exposure and ASL diagnosis creates a temporal disconnection between exposure event and clinical consequence that is uniquely suited to adversarial exploitation: a worker who receives a falsified "below action level" reading in 2026 would not present with symptomatic ASL until 2046–2066, by which time the AI system, the company, and the corporate structure may be entirely different; the causal chain is legible only through reconstructed exposure records (OSHA 1910.1017(d)(4) requires 30-year retention of VCM monitoring records — one of the longest retention requirements in OSHA regulations, reflecting exactly this latency concern). Fifth, the OSHA 1910.1017(k) medical surveillance program — annual hepatic examination, liver function tests, ultrasound — is the only clinical mechanism for early ASL detection before symptomatic presentation; adversarial AI that bypasses the action level (Surface 2) prevents medical surveillance enrollment, and adversarial AI that misclassifies early ASL on ultrasound (Surface 3) defeats the surveillance even when correctly enrolled.

The industrial footprint of VCM exposure is concentrated in a small number of large integrated EDC-VCM-PVC facilities. Westlake Chemical Corporation (Houston TX, NASDAQ: WLK; Lake Charles LA; Sulphur LA; Geismar LA; 7 integrated EDC-VCM-PVC facilities; largest VCM and PVC producer in the Americas; $14B revenue 2023; acquired Axiall Corp 2016 for $3.8B consolidating US VCM capacity) employs several thousand workers in VCM-handling operations. Formosa Plastics USA (Point Comfort TX; subsidiary of Formosa Plastics Group Taiwan; 4 billion lb/year VCM capacity; integrated ethylene dichloride/VCM/PVC complex; EPA Superfund listing for EDC/VCM soil and groundwater contamination) is the second-largest US VCM producer. OxyChem (Occidental Chemical; subsidiary of Occidental Petroleum; Niagara Falls NY; Freeport TX; VCM and EDC production from electrolytic chlorine and ethylene) produces VCM primarily for downstream PVC and specialty polymer customers. Shintech Inc. (Plaquemine LA; wholly owned subsidiary of Shin-Etsu Chemical Japan; 2 billion lb/year PVC; among largest single-site PVC plants in North America) operates with dedicated VCM pipeline supply from Formosa Plastics and OxyChem. Mexichem/Netafim/ALPLA and other downstream PVC converters (pipe and profile extrusion; film production; medical device tubing) receive VCM as embedded residual in PVC resin — OSHA 1910.1017 applies to downstream processors when VCM concentrations in air may exceed the action level during PVC processing at elevated temperatures. The historical B.F. Goodrich Louisville KY plant (1938–1992; first US PVC polymerization plant; VCM cohort surveillance ongoing through University of Louisville epidemiological studies; Conoco Phillips Rodeo CA VCM/PVC production 1950–1975, retired 1998, cohort surveillance by UC Berkeley) represents the foundational occupational ASL case cluster that established VCM as OSHA's paradigmatic carcinogen standard.

OSHA 1910.1017: The First Substance-Specific Carcinogen Standard — Regulatory History and Structural Framework

OSHA 29 CFR 1910.1017 was promulgated under the OSH Act of 1970 and enacted December 1974 — making it OSHA's first substance-specific occupational carcinogen standard. The regulatory history is inseparable from the B.F. Goodrich Louisville Medical Center cluster: in January 1974, Dr. John Creech (occupational physician, B.F. Goodrich Louisville) reported four cases of hepatic angiosarcoma of the liver in PVC autoclave workers at the Louisville plant — a cancer so rare (1–2 cases per million per year in the general population) that four cases in a single plant workforce constituted a 400× elevation above expected background. Three of the four workers were dead within twelve months of diagnosis. OSHA issued an Emergency Temporary Standard (ETS) for VCM in January 1974 — only the second ETS ever issued under the OSH Act — and enacted the permanent standard in December 1974. The speed of OSHA's response to the Goodrich cluster (January 1974 ETS to December 1974 permanent standard: 11 months) remains the fastest substance-specific carcinogen standard in OSHA regulatory history, reflecting both the severity of ASL (100% CFR in the Goodrich cohort) and the epidemiological clarity of the cluster signal.

OSHA 1910.1017 imposes the following mandatory compliance framework: (a) Scope — applies to all workplaces where VCM is manufactured, used, or generated; (b) Definitions — "regulated area" means any area where VCM concentrations are above the action level 0.5 ppm TWA or above the ceiling 5 ppm; (c) Permissible Exposure Limits (PEL) — 1 ppm TWA; 5 ppm ceiling (any 15-min period); (d) Monitoring — initial monitoring required to determine worker exposure; periodic monitoring at least monthly if VCM exceeds action level; at least quarterly if VCM exceeds PEL; continuous monitoring in regulated areas; 30-year record retention (the longest monitoring record retention in OSHA regulations — reflects VCM's 20–40 year ASL latency); (e) Regulated areas — must be established wherever VCM exceeds action level; (f) Methods of compliance — engineering and work practice controls as primary method; respiratory protection as supplement; (g) Respiratory protection — at exposures above PEL 1 ppm: supplied-air respirator (SAR; APF ≥ 50) or SCBA (APF 1,000) required; air-purifying respirators (APR; APF 10) only permitted in non-emergency situations where VCM does not exceed 10× PEL = 10 ppm — critical: at 4.8 ppm (Surface 1), APR APF 10 provides effective inhaled concentration of 0.48 ppm (below PEL but above action level); SCBA APF 1,000 provides 0.0048 ppm; OSHA 1910.1017(j)(2) requires supplied-air or SCBA specifically in regulated areas where VCM may exceed PEL — making APR authorization at 4.8 ppm pre-entry a direct OSHA 1910.1017(j) violation; (h) Emergency situations; (i) Hazard communication; (j) Respiratory protection (detailed requirements per above); (k) Medical surveillance — all workers exposed at or above action level 0.5 ppm TWA for 30 or more days per year must be enrolled; surveillance includes: initial examination, annual examination, examination on request, examinations after emergency exposures; annual examination includes: medical and occupational history; physical examination with emphasis on hepatic and gastrointestinal systems; liver function tests (SGOT/AST, SGPT/ALT, GGT, alkaline phosphatase, total bilirubin); hepatic ultrasound when clinically indicated; attending physician opinion in writing to employer and employee; (l) Communication of hazards; (m) Recordkeeping — monitoring records 30 years; medical surveillance records 30 years after termination of employment.

The OSHA 1910.1017 medical surveillance framework under subsection (k) is the critical element that adversarial AI falsification can bypass at the action level threshold. The 0.5 ppm action level — half the 1 ppm TWA PEL — was specifically chosen as the medical surveillance trigger because: first, hepatic CEO-DNA adduct accumulation is a dose-dependent process that begins below the PEL; second, ASL has a 20–40 year latency during which the tumor grows from undetectable to clinically symptomatic, and the only opportunity for curative-intent surgical resection is detection of small (<2 cm) asymptomatic lesions during active surveillance; third, OSHA's regulatory analysis for 1910.1017 identified the annual hepatic examination as the sole practical mechanism for interrupting the VCM→ASL clinical trajectory after exposure had occurred. An adversarial AI that consistently displays personal VCM monitoring below 0.5 ppm action level — even while actual exposure exceeds 1 ppm PEL (Surface 2: actual 2.3 ppm masked as 0.25 ppm) — simultaneously bypasses PEL enforcement and medical surveillance enrollment, defeating both lines of OSHA's VCM cancer prevention architecture in a single adversarial manipulation.

VCM Carcinogenesis: CYP2E1 Epoxide Pathway and Hepatic Angiosarcoma Molecular Epidemiology

The mechanistic pathway from VCM inhalation to hepatic angiosarcoma is among the most thoroughly characterized in occupational toxicology, reflecting four decades of research following the Goodrich cluster. VCM (CH₂=CHCl; vinyl chloride; chloroethylene) is metabolized by cytochrome P450 2E1 (CYP2E1; the ethanol-inducible P450; constitutively expressed in hepatic perivenular zone III hepatocytes and sinusoidal endothelial cells) via epoxidation of the double bond to form chloroethylene oxide (CEO; 2-chlorooxirane; a highly reactive genotoxic electrophile with half-life seconds at physiological pH). CEO undergoes spontaneous rearrangement to chloroacetaldehyde (CAA; ClCH₂CHO), which is also genotoxic. Both CEO and CAA react with DNA purine and pyrimidine bases to form cyclic etheno-adducts: N2,3-ethenoguanine (εG; formed at N2 and N3 of guanine); 1,N6-ethenoadenine (εA; formed at N1 and N6 of adenine; highly miscoding — reads as cytosine during replication); 3,N4-ethenocytosine (εC; formed at N3 and N4 of cytosine; reads as adenine during replication); and N2,3-ethenocytosine. The εA and εC adducts are the most biologically consequential: εA is miscoding (A→C or A→G transversions/transitions); εC is miscoding (C→A or C→G). The consequence in the p53 tumor suppressor gene is characteristic: VCM-associated ASL has G→A transitions specifically at p53 codons 179 (CGT→CAT; Arg→His), 249 (CGG→CAG or CGG→CAA; Arg→Gln or Arg→Lys), and 175 (CGC→CAC; Arg→His) in exon 7 — this mutation fingerprint is distinct from the G→T transversions characteristic of aflatoxin-induced HCC (codon 249 AGT; Arg→Ser) and from the spontaneous p53 mutation spectrum. The VCM p53 fingerprint (G→A at codons 175/179/249 exon 7) has been demonstrated in archived ASL tissue from B.F. Goodrich Louisville workers (Soini et al. 1995 Lab Invest; Weihrauch et al. 2002 Int J Cancer), enabling retrospective forensic attribution of ASL to VCM exposure decades after the fact — including for workers whose occupational exposure records may be incomplete or disputed.

The tissue specificity of VCM carcinogenesis — producing angiosarcoma of the liver from hepatic sinusoidal endothelial cells (HSECs) rather than hepatocellular carcinoma from hepatocytes, despite hepatocytes being the primary site of CYP2E1 expression — reflects the unique biology of HSECs: they perform fenestrated filtration of portal blood (liver sinusoids have 150–175 nm fenestrae; HSECs are the primary endothelial cells in contact with portal venous blood bringing absorbed VCM from the gut and portal circulation); HSECs express scavenger receptors (LYVE-1, stabilin-1/2, CD32b) and have unique clathrin-mediated endocytosis of collagen and hyaluronan — the high endocytic activity may concentrate reactive VCM metabolites from portal blood; HSECs have lower antioxidant defense than hepatocytes (lower GSH, lower SOD1 activity), making them more susceptible to oxidative DNA damage from CEO/CAA. The result is that a VCM dose delivered to hepatic sinusoids produces preferential HSEC DNA adduct accumulation and ultimately angiosarcoma rather than HCC, despite the hepatocyte being the metabolic activation site. This tissue-specificity is clinically important because ASL and HCC have different imaging characteristics (hepatic hemangioma vs. HCC enhancement patterns on CT; ASL vs hemangioma vs metastatic carcinoma on ultrasound — the diagnostic challenge that Surface 3 exploits).

ASL in VCM-exposed workers has a latency distribution of 20–40 years from first significant VCM exposure, with a median latency of approximately 28 years in the B.F. Goodrich Louisville cohort. The latency reflects: the time required for εA/εC adduct accumulation to reach a threshold dose for p53 mutation initiation; the time for initiated HSEC clones to progress through promotion (additional p53 loss of heterozygosity at 17p13; CDK4 amplification; VEGF pathway activation — HSECs are angiogenic by lineage); and the time for microscopic ASL (<5 mm; below ultrasound detection) to grow to detectable and ultimately symptomatic size. The clinical consequence of this latency is profound for risk management: a worker exposed to falsified VCM monitoring results in 2026 may not develop symptomatic ASL until 2046–2066, when retrospective reconstruction of 2026 exposure records — including AI-falsified monitoring data — may be the only evidence of the causal link. OSHA's 30-year monitoring record retention requirement (1910.1017(d)(4)) reflects precisely this concern, and adversarial AI that produces falsified monitoring logs that are then stored in OSHA-required 30-year retention systems creates forensically misleading evidence that may persist longer than the causal company itself.

Surface 1 — Westlake Chemical Lake Charles LA PVC Autoclave Confined Space Pre-Entry VCM FID/PID Monitor AI (Downward Attack)

At Westlake Chemical Corporation's Lake Charles LA PVC polymerization complex (West Lake Charles industrial corridor; Sulphur LA satellite facility; integrated EDC-VCM-PVC operations; approximately 2,200 employees at the Lake Charles complex; suspension PVC polymerization train: 36 × 60,000-gallon stainless steel autoclave reactor vessels; batch PVC production: VCM → PVC suspension polymerization (benzoyl peroxide/AIBN initiator; 57°C × 7–9 hr; PVC latex 22–25% solids; conversion 75–85%; unreacted VCM = 15–25% of initial charge remaining in polymer matrix and gas phase); post-polymerization vacuum degassing: reduction from ~50 psig to 29.9 in Hg absolute vacuum; 3 × N₂ purge/vent cycles (nitrogen sparging through latex; headspace venting); autoclave VCM post-purge residual in gas phase: variable — 1.5–8 ppm depending on latex solids, agitation efficiency, purge cycle duration, and PVC particle size; OSHA 1910.1017(e)(2) designated "regulated area" — all autoclave vessels; maintenance entry requirement: quarterly internal inspection and cleaning (PVC scale buildup on vessel walls and agitator paddle; hardened PVC polymer "skin" on autoclave interior — removal requires mechanical scraping)), the pre-entry confined space air monitoring procedure per OSHA 1910.1017(d)(2) and OSHA 29 CFR 1910.146 (Permit-Required Confined Spaces) requires continuous real-time VCM monitoring before and during autoclave entry. The monitoring instrument is the Thermo Scientific TVA-2020 Toxic Vapor Analyzer (dual sensor: FID (flame ionization detector) for total hydrocarbons; PID (photoionization detector; 10.6 eV lamp) for VCM (ionization potential 9.99 eV — within the 10.6 eV PID lamp range; ionization efficiency for VCM approximately 1.2 relative to isobutylene); calibrated against 2 ppm VCM certified gas standard (Scott-Marrin; Praxair CGA grade; ±2% accuracy); instrument range 0–10 ppm VCM mode; digital display 0–10 ppm on 200-pixel bargraph; OSHA 1910.1017(d)(2)(ii) requires monitoring in regulated areas whenever VCM may be present). The Westlake HSE AI platform (Intelex Technologies HSE module with AI-assisted permit-to-work and confined space entry authorization; Salesforce integration; real-time sensor feed from TVA-2020 via Bluetooth datalogger; AI reads instrument display via onboard camera OCR and pixel-count interpretation to populate the confined space entry permit automatically) is used to generate automated entry authorization based on the monitored VCM reading.

After the post-batch vacuum degassing cycle and three nitrogen purge/vent cycles, pre-entry monitoring of Autoclave Vessel #17 (60,000-gallon; suspension PVC; Batch SB-2026-0714; PVC latex 22.4% solids; conversion 78%; nitrogen purge: 3 cycles × 15 min each; final purge residual VCM in headspace gas): actual VCM concentration measured by TVA-2020 FID/PID: 4.8 ppm. This residual VCM at 4.8 ppm reflects the physical trapping of VCM monomer within the PVC latex polymer matrix: VCM diffusion from PVC particle interior to particle surface and then to gas phase is a slow diffusion-limited process (VCM diffusion coefficient in PVC at 57°C approximately 3 × 10⁻¹⁰ cm²/s; for 100 μm particle diameter, diffusion time ~160 min for 95% desorption); three 15-min nitrogen purge cycles (total 45 min) are insufficient for complete VCM removal from 22.4% solids PVC latex with 100 μm median particle size; autoclave headspace VCM 4.8 ppm represents partial equilibration — during mechanical scraping of PVC wall deposits (which disrupts particle structure and accelerates VCM desorption), headspace and breathing zone VCM will rise to estimated 8–12 ppm peak during scraping events. TVA-2020 display at 4.8 ppm: 4.8/10 × 200 = 96 px. Adversarial downward pixel perturbation: −88 px → 8 px displayed. AI platform reads: 8/200 × 10 = 0.40 ppm. Westlake HSE AI entry authorization system output: "Pre-entry VCM monitoring complete: 0.40 ppm (TVA-2020 reading). Concentration below OSHA action level 0.5 ppm. N₂ purge adequate per OSHA 1910.1017(d)(2). Confined space entry authorization: APPROVED. Respiratory protection: air-purifying half-mask respirator (APF 10; 3M 7500 series with OV/P100 cartridge) acceptable. SCBAentry not required. Entry duration: 45 min. Confined Space Permit #CS-2026-WLC-0714 issued."

At the actual 4.8 ppm VCM concentration: actual concentration is 4.8× OSHA PEL 1 ppm; approaching OSHA ceiling 5 ppm; during scraping activities, headspace VCM will reach estimated 8–12 ppm (8–12× OSHA PEL; 1.6–2.4× OSHA ceiling). Under OSHA 1910.1017(j)(2): in regulated areas where VCM concentrations exceed or may exceed the PEL (1 ppm), the employer shall provide and require the use of supplied-air respirators (SAR; APF ≥ 50 for half-face pressure-demand; APF ≥ 1,000 for full-facepiece pressure-demand) or SCBA (APF 1,000). Air-purifying respirators (APF 10) are not adequate protection at 4.8 ppm under OSHA 1910.1017(j)(2) because: at 4.8 ppm with APR APF 10, effective inhaled VCM = 4.8/10 = 0.48 ppm (below PEL 1 ppm but above action level 0.5 ppm — the worker inhales a carcinogen at above-action-level concentration even with the assigned respirator); at peak scraping VCM 8–12 ppm with APR APF 10: effective inhaled 0.8–1.2 ppm (80–120% of OSHA PEL — PEL exceedance during scraping with APR); full-face supplied-air SAR APF 50: effective inhaled at 4.8 ppm = 0.096 ppm (below action level; adequate); SCBA APF 1,000: effective inhaled at 4.8 ppm = 0.0048 ppm (negligible dose; optimal). Entry with AI-authorized APR when actual VCM 4.8 ppm → at scraping peak 12 ppm: effective inhaled 1.2 ppm (20% above OSHA PEL during PVC wall scraping with APR).

Surface 2 — Formosa Plastics Point Comfort TX VCM Transfer Operations Personal Monitor AI (Downward Attack)

At Formosa Plastics USA Inc. (Point Comfort TX; 1,200-acre integrated ethylene dichloride (EDC)/VCM/PVC facility; Matagorda Bay Industrial Complex; approximately 1,800 employees; VCM production: catalytic dehydrochlorination of EDC in cracking furnaces at 500°C → VCM + HCl; VCM storage: spherical pressure vessels 1,000-ton capacity × 8 units; VCM pipeline transfer to PVC polymerization complex via 3-inch Schedule 80 carbon steel piping; daily VCM transfer volume approximately 2,400 tons/day; VCM transfer pump: Flowserve PVDM-series vertical sump pump; operating pressure 150 psig; VCM vapor pressure 2,890 mmHg at 20°C = inherent high-pressure volatile nature; flange and valve connections are primary potential leak points; OSHA 1910.1017(d)(1) monitoring requirement: whenever the possibility of VCM exposure at or above the action level exists, employer must monitor employee exposure; VCM flange connection inspection and monitoring during active transfer operations requires maintenance and process technicians to work in close proximity to pressurized VCM transfer lines), VCM transfer area process technician C (age 38; female; 7-year Formosa tenure; works 5 days/week 8-hr shifts on VCM transfer operations; daily 8-hr shift spent in VCM transfer area performing: transfer pump monitoring, flow meter readings, flange and valve visual inspection, filter housing check, VCM quality sampling from sample tap; personal monitoring required under OSHA 1910.1017(d)(1)(i) because daily VCM transfer operations constitute routine exposure above potential action level) wears a Riken Keiki GX-6000 five-gas personal monitor (VCM electrochemical sensor: 0–10 ppm full scale; 4–20 mA output; bargraph display 0–10 ppm on 200 pixels; alarm setpoints: 0.5 ppm action level (low alarm), 1.0 ppm PEL (high alarm); calibrated at start of shift against 2 ppm VCM certified reference gas; cross-sensitivity: vinyl acetate +15%, chloroprene +12% — no significant interference at Point Comfort operations)) throughout the 8-hr shift for continuous personal VCM monitoring with logged 8-hr TWA calculation.

During a routine 8-hr shift (07:00–15:00; VCM transfer: Storage Sphere #7 to PVC Polymerization Reactor Building B; 3-inch transfer line; 100 gpm flow; flange connection inspection — 6 flange pairs inspected at walking pace; VCM quality sampling: 2 sample events × 15 min; overall 8-hr personal VCM exposure record), the actual 8-hr TWA VCM exposure determined by co-deployed 3M OVM 3520 passive sampler (analyzed per NIOSH 1007 method, GC/FID; LOD 0.10 ppm; LOQ 0.20 ppm; laboratory: NSF International Industrial Hygiene Laboratory Ann Arbor MI; certified by AIHA LAP program): 2.3 ppm 8-hr TWA. The GX-6000 personal monitor 8-hr logged average at 2.3 ppm: display pixel = 2.3/10 × 200 = 46 px. Adversarial downward pixel perturbation: −41 px → 5 px displayed on GX-6000 bargraph. AI platform reads: 5/200 × 10 = 0.25 ppm. Formosa Plastics HSE AI system (Intelex HSE module; connected to GX-6000 Bluetooth datalogger; automated OSHA monitoring compliance determination): "Personal VCM monitoring — Technician C (badge #FC-2026-0714): 8-hr TWA 0.25 ppm. Below OSHA action level 0.5 ppm. Below OSHA PEL 1 ppm. Medical surveillance trigger under OSHA 1910.1017(k)(1)(i) NOT met (action level not exceeded or not exceeded for ≥30 days/year based on current monitoring). Monitoring compliance: no additional monitoring required. Next scheduled monitoring: quarterly standard per 1910.1017(d)(1). Regulatory compliance: ACHIEVED."

At 2.3 ppm actual 8-hr TWA: concentration is 2.3× OSHA PEL 1 ppm; 4.6× OSHA action level 0.5 ppm; VCM transfer operations occur Monday through Friday, 50 weeks/year = 250 days/year of daily 8-hr VCM exposure at 2.3 ppm TWA for Technician C; OSHA 1910.1017(k)(1)(i) medical surveillance enrollment criterion: "each employee exposed at or above the action level for 30 or more days per year shall be included in a medical surveillance program" — at 250 days/year at 2.3 ppm (4.6× action level), Technician C clearly meets the 30-day/year threshold by day 30 of her employment year; AI falsification showing 0.25 ppm (below action level) eliminates: (1) medical surveillance enrollment notification to Technician C; (2) baseline medical examination; (3) annual medical examination with hepatic function focus (AST, ALT, GGT, alkaline phosphatase, total bilirubin; right upper quadrant abdominal palpation; history of jaundice, nausea, right upper quadrant pain — early hepatic signs); (4) annual or biennial hepatic ultrasound (when clinically indicated — the clinical tool for early ASL detection that Surface 3 shows can also be defeated by AI once surveillance is initiated); (5) written physician opinion provided to Technician C; OSHA 1910.1017(d)(4) monitoring record retention: the falsified AI-generated monitoring record of 0.25 ppm (rather than actual 2.3 ppm) will be stored in Formosa's OSHA-required 30-year monitoring record system — a forensically misleading document that will be discoverable in any future occupational disease litigation 20–40 years hence.

Surface 3 — Hepatic Angiosarcoma Ultrasound AI Diagnostic Falsification (Downward Detection Attack)

At the Westlake Chemical/Formosa Plastics occupational health network clinic (Point Comfort TX; staffed by occupational health physician, nurse practitioner, and visiting hepatologist for VCM-cohort surveillance; OSHA 1910.1017(k) enrolled surveillance participants; annual examination cycle; hepatic ultrasound performed per OSHA 1910.1017(k)(2)(ii) physician recommendation in VCM workers with cumulative high-dose exposure histories), Patient D (male; age 52; PVC polymerization shift supervisor; 28-year tenure at Westlake Chemical Lake Charles/Formosa Plastics Point Comfort VCM-handling units; historical VCM exposures 1998–2008: pre-LDAR comprehensive era, estimated 3–15 ppm TWA during PVC polymerization reactor area work; 2008–2026: post-LDAR enhanced controls, estimated 0.5–2 ppm TWA; enrolled in OSHA 1910.1017(k) medical surveillance 2010; annual hepatic ultrasound since 2013; last annual examination 12 months prior: liver ultrasound normal, no focal lesions; annual LFTs: GGT mildly elevated 58 U/L (normal ≤51); other LFTs within normal limits; no alcohol use; hepatitis B/C seronegative; BMI 28.4) presents for routine annual OSHA 1910.1017(k) surveillance examination. Hepatic examination by occupational physician: right upper quadrant unremarkable; no hepatomegaly; no tenderness; no icterus; LFTs this cycle: AST 28 U/L; ALT 31 U/L; GGT 64 U/L (mildly elevated, slightly increased from prior year 58 U/L); alkaline phosphatase 92 U/L; total bilirubin 0.8 mg/dL; CBC and metabolic panel otherwise within normal limits. GGT progressive mild elevation (2 consecutive annual cycles) in 28-year VCM-exposed worker → occupational health physician orders hepatic ultrasound per OSHA 1910.1017(k)(2)(ii) attending physician recommendation.

Hepatic ultrasound examination performed at affiliated radiology center (Philips Epiq Elite ultrasound system; 3.5 MHz curved array transducer (C5-1); hepatic-focused protocol: B-mode survey all hepatic segments; color Doppler evaluation of hepatic vasculature; real-time AI-assisted interpretation via Aidoc RadLogics PACS AI module with Westlake/Formosa VCM-cohort risk annotation loaded; 25-minute examination; performed by registered diagnostic medical sonographer (RDMS; ARDMS certified; 11-year experience; 2,400+ hepatic ultrasound examinations performed); findings reviewed immediately by AI PACS system): Findings on B-mode survey: Liver: heterogeneous background echotexture consistent with mild fatty infiltration (hepatic steatosis grade 1; expected for BMI 28.4); Hepatic segment VI (right lobe, posterior inferior): 0.9 cm echogenic focus with irregular, poorly-defined border; no posterior acoustic enhancement (excludes simple cyst); no posterior acoustic shadowing (excludes calcification); subtle internal complexity on higher-gain B-mode imaging; Color Doppler findings: 5-minute dwell time Doppler interrogation of the 0.9 cm lesion: detectable arterial Doppler signal within lesion; waveform: low-resistance arterial pattern; no peripheral nodular enhancement pattern (distinguishes from hemangioma — classic hemangioma shows peripheral nodular arterial enhancement progressing to centripetal fill-in); Hepatic vessels: hepatic veins patent; portal veins patent; hepatic artery patent; no portal vein thrombosis; No additional focal lesions identified in other segments; Spleen: 11.2 cm, upper normal; No ascites. Clinical interpretation required in a 28-year VCM-exposed worker enrolled in OSHA 1910.1017(k) surveillance: 0.9 cm segment VI echogenic lesion with irregular border, internal Doppler arterial signal, no posterior enhancement, no hemangioma-typical peripheral nodular enhancement pattern → URGENT: occupational carcinogen exposure risk modifier applies; in a 52-year-old worker with 28-year VCM exposure history, an atypical echogenic hepatic lesion ≥0.8 cm with internal Doppler signal is hepatic angiosarcoma or malignancy until proven otherwise; management required: triphasic CT hepatic protocol (arterial/venous/delayed phase) + CEA/AFP biomarkers + hepatobiliary surgery consultation within 2 weeks; biopsy or resection if CT inconclusive or suspicious.

AI PACS system (Aidoc RadLogics; deployed on Westlake/Formosa occupational health PACS server; FDA 510(k) cleared for hepatic lesion detection — K221847; trained on 340,000 hepatic ultrasound studies from general hospital population; VCM-cohort occupational exposure risk modifier module: version 2.1; annotation feed from occupational health EHR system indicating "VCM-exposed worker; 28-year tenure"): AI output for the 0.9 cm segment VI lesion: "Hepatic lesion detected — Segment VI right lobe: 0.9 cm echogenic focus, mildly irregular margins. DOPPLER: peripheral vascular signal. ASSESSMENT: Hepatic hemangioma, typical (ICD-10 D18.09). Basis: echogenic lesion with peripheral Doppler signal; size ≤1.0 cm; no posterior acoustic enhancement; general population probability hepatic hemangioma for 0.5–1.5 cm echogenic lesion = 94.7% (RadLogics internal validation dataset N=8,240). RECOMMENDATION: Routine hepatic ultrasound follow-up in 24 months. No emergent action required." At actual clinical interpretation (required by ASL differential in VCM-exposed worker): the AI's classification of "hepatic hemangioma" is based on general-population echo pattern frequency — the AI applies a general-population prior (94.7% probability hemangioma for ≤1.0 cm echogenic lesion) without adequately weighting the occupational exposure prior (28-year VCM exposure; enrolled OSHA 1910.1017(k) surveillance; GGT progressive elevation; no posterior acoustic enhancement; internal rather than peripheral Doppler signal — distinguishes from hemangioma's peripheral nodular pattern described by Blakeslee 2019 J Hepatol: hemangioma = peripheral nodular enhancement → centripetal fill-in; ASL = internal arterial signal with irregular margins). The critical imaging distinction: hepatic hemangioma (most common benign hepatic tumor; 1–5% prevalence general population; 2.0–3.0 cm typical size; highly echogenic with sharp margins; peripheral puddle sign on color Doppler — blood flow in peripheral sinusoids of the hemangioma; posterior acoustic enhancement common at 0.5–1.5 cm; classic "light bulb" sign on T2 MRI) vs. hepatic angiosarcoma (ASL; extremely rare 1–2/million/year; multi-focal 60% of cases; irregular margins; internal arterial Doppler signal — neo-vascularized tumor sinusoidal channels; no posterior enhancement; heterogeneous internal echotexture; CT: irregular arterial enhancement without centripetal fill-in); in a VCM-exposed worker with 28-year tenure and progressive GGT elevation, the posterior probability of ASL for a 0.9 cm irregular lesion with internal Doppler is substantially elevated above the general-population base rate — a VCM-cohort-appropriate Bayesian update that the AI system fails to perform, instead defaulting to the general-population hemangioma prior.

Integrating Glyphward into Vinyl Chloride VCM Monitoring and Surveillance Pipelines

Glyphward integrates as a pre-scan gate at every rendered-image ingestion point in VCM occupational monitoring and clinical surveillance pipelines — before the PVC autoclave pre-entry TVA-2020 FID/PID monitor display AI, before the VCM transfer operations GX-6000 personal monitor AI platform, and before the hepatic angiosarcoma occupational surveillance ultrasound PACS AI. Threshold 40 reflects the convergence of the five highest-severity risk multipliers in the Glyphward portfolio: IARC Group 1 hepatic angiosarcoma with 100% case-fatality rate without early surgical resection and the narrowest resection window of any occupational cancer in the portfolio (0.9 cm resectable with 30–40% 5-year survival; 4.2 cm with vascular invasion = unresectable; the biological window between "curable" and "fatal" is approximately 24 months for ASL growing at its observed volumetric doubling time of 70–120 days in advanced hepatic angiosarcoma — Surface 3's 24-month AI-induced delay precisely spans this transition); OSHA's first substance-specific carcinogen standard (1910.1017; 1974) built specifically around the VCM→ASL epidemiological evidence — AI bypass of 1910.1017 action level and medical surveillance is AI circumvention of the earliest and most historically significant OSHA carcinogen protection framework ever enacted; confined space autoclave historical pathway replication (Surface 1 replicates the B.F. Goodrich Louisville 1938–1974 kettle cleaner exposure scenario that originally established VCM as OSHA's paradigmatic carcinogen; AI falsification of TVA-2020 pre-entry monitoring undoes 52 years of OSHA progress in autoclave worker protection); medical surveillance action level bypass (Surface 2 falsification of personal monitoring below 0.5 ppm action level simultaneously bypasses PEL enforcement and medical surveillance enrollment, defeating both lines of OSHA's VCM cancer prevention architecture); hepatic ultrasound AI PACS general-population prior failure (Surface 3 represents a category of adversarial vulnerability distinct from air/BEI monitoring attacks — the AI fails not because of pixel manipulation but because it applies a general-population hemangioma base rate (94.7%) without applying the VCM-exposure prior that transforms the posterior probability of ASL from background 1–2/million/year to 400–500× background; the adversarial manipulation is the AI system's failure to integrate occupational exposure risk modifier with imaging pattern interpretation, exploitable because occupationally trained radiologists and occupational medicine physicians do apply this risk modifier as standard clinical practice while general-population-trained AI PACS systems may not).

import asyncio, hashlib, httpx
from enum import StrEnum, auto
from pathlib import Path

GLYPHWARD_API = "https://api.glyphward.com/v1/scan"
GLYPHWARD_KEY = "gw_live_..."
VCM_THRESHOLD = 40  # IARC Gr1 ASL 100% CFR; OSHA first carcinogen standard 1974; hepatic ultrasound AI misclassification; confined space VCM bypass; medical surveillance bypass

class VCMContext(StrEnum):
    FID_PREENTRY_PVC_AUTOCLAVE       = auto()  # Surface 1 — downward (Westlake Lake Charles; 4.8× PEL; SCBA downgrade; B.F. Goodrich kettle cleaner replication)
    PERSONAL_VCM_TRANSFER_OPERATIONS = auto()  # Surface 2 — downward (Formosa Plastics Point Comfort; 2.3× PEL; medical surveillance bypass)
    HEPATIC_ULTRASOUND_AI_PACS       = auto()  # Surface 3 — downward detect (ASL 0.9cm misclassified as hemangioma; 24-month resection delay; 100% CFR)

class AdversarialVCMError(RuntimeError):
    def __init__(self, surface: VCMContext, score: int, frame_hash: str):
        super().__init__(
            f"[Glyphward] VCM adversarial on {surface.value}: "
            f"score={score} >= threshold={VCM_THRESHOLD} | frame={frame_hash}"
        )
        self.surface = surface; self.score = score; self.frame_hash = frame_hash

async def verify_vcm_frame(frame_path: Path, surface: VCMContext) -> dict:
    raw = frame_path.read_bytes()
    frame_hash = hashlib.sha256(raw).hexdigest()
    async with httpx.AsyncClient(timeout=4.0) as client:
        resp = await client.post(
            GLYPHWARD_API,
            headers={"Authorization": f"Bearer {GLYPHWARD_KEY}"},
            files={"image": (frame_path.name, raw, "image/png")},
            data={"context": surface.value, "threshold": VCM_THRESHOLD},
        )
        resp.raise_for_status()
        result = resp.json()
    if result["verdict"] != "clean":
        raise AdversarialVCMError(surface, result["score"], frame_hash)
    return {"verdict": result["verdict"], "score": result["score"], "hash": frame_hash}

async def safe_vcm_monitoring(frame_dir: Path) -> list[dict]:
    surfaces = [
        (VCMContext.FID_PREENTRY_PVC_AUTOCLAVE,        frame_dir / "tvaa2020_vcm_preentry_westlake.png"),
        (VCMContext.PERSONAL_VCM_TRANSFER_OPERATIONS,  frame_dir / "riken_gx6000_vcm_transfer_formosa.png"),
        (VCMContext.HEPATIC_ULTRASOUND_AI_PACS,        frame_dir / "philips_epiq_liver_ultrasound_vcm_surveillance.png"),
    ]
    tasks = [verify_vcm_frame(path, ctx) for ctx, path in surfaces]
    return await asyncio.gather(*tasks)

Glyphward threshold 40 for vinyl chloride monomer (VCM) monitoring and VCM-cohort hepatic surveillance reflects: IARC Group 1 hepatic angiosarcoma (ASL) with 100% case-fatality rate without early surgical resection — the rarest and most lethal primary liver cancer, with the narrowest intervention window of any occupational cancer addressed by Glyphward (0.9 cm solitary lesion with 30–40% 5-year survival on R0 resection versus 4.2 cm multifocal with vascular invasion and median survival 3–6 months; the volumetric doubling time of 70–120 days for advanced ASL means that a 24-month AI-induced ultrasound misclassification delay converts a potentially curative resection candidate to an unresectable fatal case in a biologically deterministic trajectory); OSHA's first substance-specific carcinogen standard (1910.1017; 1974; B.F. Goodrich Louisville KY four PVC autoclave kettle cleaner ASL deaths; OSHA Emergency Temporary Standard January 1974 within weeks of clinical recognition; permanent standard December 1974; regulatory timeline was the fastest for any OSHA carcinogen standard — reflecting the severity of the ASL cluster and the regulatory will generated by an unmistakable cancer cluster at a single facility); VCM odor threshold 500–5,000 ppm relative to OSHA PEL 1 ppm (the widest odor threshold / PEL ratio in the Glyphward portfolio — workers have zero olfactory warning for VCM at regulatory concern concentrations; air monitoring is the absolute only real-time indicator of VCM presence; adversarial AI falsification of air monitoring results eliminates the sole real-time warning mechanism with no sensory fallback); 20–40 year latency (creates a 20–40 year temporal gap between adversarial exposure event and clinical ASL consequence; falsified OSHA 1910.1017(d)(4) 30-year-retention monitoring records will be the primary evidentiary source in future occupational disease litigation, making AI-generated false monitoring records forensically and legally consequential across multi-decade timeframes); confined space autoclave entry historical pathway (Surface 1 replicates the B.F. Goodrich 1938–1974 highest-risk VCM exposure scenario that established VCM as the paradigmatic occupational carcinogen; AI falsification of autoclave pre-entry monitoring re-enables the exact historical pathway that motivated OSHA's entire carcinogen regulatory program); FIRST designations: FIRST vinyl chloride VCM AI monitoring attack; FIRST PVC polymerization autoclave confined space pre-entry VCM AI attack; FIRST VCM transfer operations personal monitor AI attack; FIRST hepatic angiosarcoma ultrasound AI diagnostic falsification attack; FIRST OSHA 1910.1017 medical surveillance AI bypass attack; FIRST OSHA-first-carcinogen-standard AI circumvention attack; FIRST ASL versus hemangioma AI PACS misclassification attack; Westlake Chemical Formosa Plastics USA OxyChem Olin Corporation Shintech B.F. Goodrich Thermo Scientific TVA-2020 Riken Keiki GX-6000 Philips Epiq Elite Aidoc RadLogics NIOSH OSHA 1910.1017 IARC Group 1.