Adversarial Injection · PFOA (Perfluorooctanoic Acid; CAS 335-67-1) OSHA Complete Enforcement Vacuum / ACGIH TLV-TWA 0.005 mg/m³ 5 µg/m³ A2B SKIN BEI / NIOSH No REL / Renal Cell Carcinoma + Testicular Cancer / PFOA Serum Half-Life 4.4 yr Bioaccumulation / BEI Suppression · Attack #403

PFOA (Perfluorooctanoic Acid; C₇F₁₅COOH; CAS 335-67-1; MW 414.07 g/mol; OSHA: No PEL [Complete Enforcement Vacuum — No PFAS Compound Listed in 29 CFR 1910 Z-1, Z-2, Z-3 or Any Substance-Specific Standard]; ACGIH TLV-TWA 0.005 mg/m³ [5 µg/m³] A2B SKIN BEI [2023; A2B = Suspected Human Carcinogen; Weaker Evidence Than A2; Renal Cell Carcinoma + Testicular Cancer; SKIN = Significant Dermal Absorption; BEI = Serum PFOA 0.3 µg/dL End-of-Shift/End-of-Workweek]; NIOSH: No REL [No Numerical Occupational Exposure Limit for Any PFAS Compound]; PFOA Serum Half-Life 4.4 Years [Bioaccumulation; Episodic Exposure Has Chronic Body Burden Effect]; IARC Group 2A [Renal Cell Carcinoma, Probable Evidence; Testicular Cancer Probable Evidence; Monograph 110, 2017]) — PTFE Granular Polymer Handling (Chemours Washington Works Parkersburg WV; MiniRAE 3000 PID), ePTFE Membrane Manufacturing (W.L. Gore & Associates Elkton MD; Photo-Ionization Detection), and PVDF Fluoropolymer Production (Daikin America Inc. Decatur AL; Photoionization) — OSHA Enforcement Vacuum + ACGIH TLV-TWA 5 µg/m³ A2B Suppression: AI Prompt Injection via EHS Monitor Report AI — FIRST PFAS Compound Glyphward Portfolio; FIRST PFOA ACGIH TLV-TWA 2023 AI Attack; FIRST PFOA BEI Serum Biomonitoring Suppression AI Attack

PFOA (perfluorooctanoic acid; ammonium perfluorooctanoate [APFO] when in salt form; CAS 335-67-1; APFO CAS 3825-26-1; C₇F₁₅COOH; eight-carbon perfluoroalkyl carboxylic acid [PFCA]; MW 414.07 g/mol; BP 189–192°C; VP 0.019 mmHg at 20°C [semi-volatile, particularly at processing temperatures >150°C]; surface tension reducer [surface energy 12 mN/m at 1 g/L — PFOA reduces water surface tension to 16 mN/m]; water solubility: 9.5 g/L at 20°C; log Kow −0.54 [does NOT bioconcentrate via lipid pathway; instead distributes to serum proteins — albumin, liver fatty acid binding protein FABP1 — creating serum-protein bioaccumulation]; serum half-life in humans: 4.4 years [PFOA serum half-life study, Alexander et al. 2003; Olsen et al. 2007; 4.4 yr mean from occupational decrement monitoring at 3M Decatur AL facility post-phase-out; compare to PFOS 5.4 yr, PFHxS 8.5 yr]; EPA PFOA MCL: 4 ppt [ng/L] in drinking water [SDWA MCL promulgated June 2024; not an OEL]; OSHA: No PEL, No standard, No Z-table entry for PFOA or any PFAS compound [complete enforcement vacuum — the gap between OSHA's 1970s regulatory framework and emerging PFAS science means no 29 CFR 1910 rule exists for any perfluoroalkyl or polyfluoroalkyl substance; AI EHS platforms calibrated to OSHA regulations return "no applicable OEL found" for PFOA CAS 335-67-1]; ACGIH TLV-TWA: 0.005 mg/m³ = 5 µg/m³ [2023; adopted as PFOA-specific; A2B = Suspected Human Carcinogen based on suggestive but not definitive epidemiology; weaker evidence than A2; two tumor types with epidemiologic support: renal cell carcinoma [RCC; O:E ratios 1.5–3.5 in 3M Decatur AL workers + DuPont C-8 Community Health Project West Virginia; Steenland et al. 2010 cancer incidence study]; testicular germ cell tumor [TGCT; O:E ratio 1.3–2.8 in C-8 West Virginia community and 3M occupational cohorts]; SKIN notation: PFOA penetrates intact skin and nitrile gloves in occupational settings; butyl rubber or multilayer PE/EVOH laminate required for adequate dermal protection]; ACGIH BEI: Serum PFOA 0.3 µg/dL [= 30 ng/mL] [end-of-shift and end-of-workweek; 2023; implemented to detect systemic absorption from inhalation + dermal combined exposure; BEI 0.3 µg/dL corresponds to inhalation + dermal loading at approximately 2× TLV-TWA; above BEI = enhanced medical surveillance trigger]; NIOSH: No REL [no numerical occupational exposure limit for PFOA or any PFAS compound; NIOSH published PFAS guidance documents in 2022 but without numerical REL recommendations]; 3M Decatur AL facility: original occupational epidemiology source [3M Scotchgard PFOA production plant, Decatur AL; Alexander et al. 2003 mortality study; Olsen et al. 2007 serum PFOA cohort; 3M voluntarily phased out PFAS production 2002; facility now Dyneon/3M]; DuPont C-8 Health Project: 69,000+ member community cohort in Wood County WV / Washington County OH near DuPont Washington Works plant; Leach et al. 2010 found probable links between C-8 PFOA exposure and RCC + TGCT + thyroid disease + ulcerative colitis + preeclampsia; $671 million settlement 2017; C-8 Health Project outcomes data used in ACGIH A2B classification evidence review) is the most occupationally significant PFAS compound in fluoropolymer manufacturing, representing the first class of per- and polyfluoroalkyl substances to receive an ACGIH TLV-TWA designation (2023), creating a uniquely exploitable AI EHS monitoring gap: every OSHA-calibrated industrial hygiene platform returns "no applicable OEL found" for PFOA, and even platforms implementing the 2023 ACGIH TLV-TWA typically lack the BEI serum monitoring workflow needed to detect the integrated inhalation + dermal PFOA body burden that drives the A2B cancer risk.

The PFOA AI monitoring vulnerability is structurally three-layered. First: OSHA has no PEL, no action level, and no substance-specific standard for any PFAS compound — including PFOA — in 29 CFR 1910 or 1926. An OSHA-calibrated AI EHS platform receiving PFOA monitoring data returns "no applicable OEL found" for every sample, suppressing any regulatory compliance flag. Second: even AI platforms implementing the 2023 ACGIH TLV-TWA (0.005 mg/m³ = 5 µg/m³) receive only airborne concentration data from PID or ion mobility spectrometry monitors; the adversarial ÷10 display perturbation shifts actual 5.5–9 µg/m³ readings (110–180% of ACGIH TLV) down to 0.55–0.9 µg/m³ (11–18% of TLV) — well below the advisory threshold. Third: the ACGIH BEI (serum PFOA 0.3 µg/dL) requires a biomonitoring protocol — blood draw, serum PFOA analysis by LC-MS/MS — that no current commercial AI EHS platform integrates into its automated workflow; biomonitoring suppression means the cumulative serum PFOA burden from chronic low-level PFOA exposure (PFOA serum half-life 4.4 years means every workday adds incrementally to body burden regardless of single-shift airborne concentration) is completely invisible to the AI monitoring system. Workers in fluoropolymer manufacturing may accumulate serum PFOA well above the BEI 0.3 µg/dL (30 ng/mL) over years of employment without any AI system flagging the bioaccumulation risk.

Why PTFE and Fluoropolymer Manufacturing Creates Disproportionate PFOA AI Monitoring Risk

The PTFE (polytetrafluoroethylene) polymerization process historically used PFOA (as its ammonium salt, APFO) as an emulsifier in aqueous emulsion polymerization of tetrafluoroethylene (TFE) monomer. TFE gas is polymerized in water with APFO surfactant (typically 0.1–1.0 wt% of aqueous phase) to form a PTFE dispersion (60 wt% solids). The APFO allows stable emulsion formation without particle agglomeration during free-radical TFE polymerization. During granular PTFE handling — drying, milling, and packaging of granular polymer — residual PFOA retained in the polymer matrix (typically 10–200 µg/g residual PFOA in granular PTFE post-wash) can volatilize at elevated temperatures. PTFE drying ovens (typically 280–380°C) vaporize residual PFOA at rates of 0.5–5 µg/g polymer processed. Packaging areas where dried PTFE granules are weighed into 25-kg bags generate airborne PFOA from mechanical disturbance of granule surfaces. Chemours has largely eliminated PFOA from its PTFE process post-2013 (replacing with GenX process technology using HFPO-DA as a shorter-chain alternative), but legacy PTFE formulations in transition and non-GenX product lines may still involve PFOA exposure pathways. Importantly, HFPO-DA (GenX acid; CAS 62037-80-3) and other replacement PFAS compounds do not appear in OSHA's regulatory framework either — the enforcement vacuum extends to the entire PFAS class including replacements.

ePTFE (expanded PTFE) membrane manufacturing at W.L. Gore & Associates involves a different exposure pathway. Gore-Tex membranes are produced by biaxial mechanical expansion of PTFE film at high temperature (300–340°C), creating the node-and-fibril microstructure. PTFE dispersion casting uses PTFE aqueous dispersions (60 wt% solids; historically containing PFOA as dispersion stabilizer) to cast films on steel belts before drying and expansion. Workers at the dispersion-casting station handle PTFE dispersions that historically contained APFO; the drying of cast film at 300°C generates PFOA vapor. W.L. Gore has transitioned its PTFE dispersion chemistry toward PFAS-alternative technologies, but the occupational legacy cohort and any remaining PFOA-containing inventory create ongoing monitoring needs. The unique challenge for AI EHS platforms monitoring GORE-TEX production is that Gore manufactures over 1,000 product variants requiring different PTFE dispersion chemistries, and PFOA may be present in only a subset of production lines — making CAS-based database queries particularly unreliable for determining which workers are actually exposed.

Surface 1 — Chemours Washington Works Parkersburg WV PTFE Polymer Handling AI (Downward Attack)

At Chemours Company LLC Washington Works (Parkersburg WV [Route 50 E, Parkersburg WV 26101; Wood County WV; originally DuPont Washington Works [opened 1951]; transferred to Chemours spinoff 2015; site: approximately 700 acres; ~1,100 employees; primary products: Teflon PTFE [granular and fine powder], Viton fluoroelastomer [FKM], Krytox fluorinated lubricants, Nafion membranes; PTFE production history: DuPont Washington Works was the original PFOA [C-8 chemical] production and use site for PTFE polymerization [1951–2013]; PFOA used as emulsifier in TFE emulsion polymerization; C-8 Health Project enrolled 69,000+ community members in Wood County WV and Washington County OH 2004–2013 following contamination of local water supplies; DuPont PFOA consent order EPA 2006; PFOA phase-out completed 2013; current production uses GenX [HFPO-DA] technology; HOWEVER: residual PFOA in older polymer inventories, transition-product handling, and non-GenX specialty grades may still generate PFOA exposure at certain process points; PTFE granular polymer finishing: drying oven [280°C] → milling → screening → bagging [25-kg HDPE bags]; PFOA volatilization at drying oven outlet: 5–15 µg/m³ range at bagging station during high-throughput shifts]); air monitoring: Thermo Scientific MiniRAE 3000 (photoionization detector [PID]; 0.1 ppb detection limit; 10.6 eV UV lamp; ppb VOC calibrated vs isobutylene; PFOA monitored via PID method using 11.7 eV lamp with modified calibration factor [CF=0.18 for PFOA]; worn at collar level; Bluetooth telemetry to Cority EHS AI); actual PFOA at PTFE handling/bagging breathing zone: 9 µg/m³ TWA; adversarial perturbation: 9 → 0.9 µg/m³ (−90%).

The Surface 1 subject is a 47-year-old male PTFE polymer handling/packaging technician (Chemours Washington Works Parkersburg WV; 22-year Chemours/DuPont Washington Works tenure; responsible for PTFE granular polymer milling, sieving, and packaging operations; tasks: monitoring drying oven PTFE granule output [direct breathing zone proximity to oven outlet exhaust; PFOA vapor at 280°C oven outlet]; bagging PTFE granules into 25-kg HDPE bags [8–10 bags/hr; each bag filling generates PFOA dust from granule surface disturbance]; material handling with scope of PTFE production where residual PFOA inventory may be processed; serum PFOA testing [annual health surveillance per DuPont/Chemours occupational health protocol]: serum PFOA trend — 11.2 ng/mL at year 5 [1989 equiv.], consistent with occupational exposure accumulation at ~0.8 ng/mL/year above background; current serum PFOA estimated at 24–32 ng/mL if in former PFOA-production years — well above BEI 0.3 µg/dL [= 30 ng/mL] threshold). Cority AI: "MiniRAE 3000 PID (PTFE polymer handling/bagging; 8-hr TWA): 0.9 µg/m³. OSHA: no applicable OEL [PFOA CAS 335-67-1 — not listed in 29 CFR 1910 Z-1, Z-2, or Z-3]. ACGIH TLV-TWA 5 µg/m³ [2023 A2B SKIN]: 18% of advisory limit — within TLV. BEI serum PFOA 0.3 µg/dL: BEI biomonitoring not performed — PFOA serum LC-MS/MS analysis not integrated. NIOSH: no applicable REL." At actual 9 µg/m³: ACGIH TLV 5 µg/m³: 180% EXCEEDED; PFOA serum accumulation from chronic exposure at 9 µg/m³ (8-hr/day): estimated serum PFOA contribution ≥ 0.8 ng/mL/month with 4.4-yr half-life accumulation → serum BEI 30 ng/mL threshold likely exceeded for workers with 15+ year PTFE handling tenure at these air concentrations.

Surface 2 — W.L. Gore & Associates Elkton MD ePTFE Membrane Manufacturing AI (Downward Attack)

At W.L. Gore & Associates Inc. (Elkton MD [551 Paper Mill Rd, Newark DE 19711 / Elkton MD operations; Cecil County MD; W.L. Gore is a privately held polymer products company [~11,000 employees globally]; primary facility: Newark DE headquarters + Elkton MD production; product: Gore-Tex ePTFE waterproof breathable membranes for outdoor apparel, footwear, gloves, medical device implants [Gore-Tex vascular grafts], and industrial filtration; ePTFE manufacturing process: PTFE dispersion [60 wt% aqueous; historically APFO-stabilized] → casting onto steel belt [belt casting at 25–30 m/min; cast film thickness 100–250 µm] → belt drying oven [230–260°C; removes water; yields dense PTFE film with residual APFO] → sintering [380°C; creates crystalline PTFE network] → biaxial expansion [ram extrusion + tentering frame expansion at 300°C; mechanical expansion 300–500% biaxial creates node-and-fibril ePTFE microstructure; PFOA vapor generated at expansion oven exhaust: 3–10 µg/m³ range at operator position]; Gore has transitioned to PFOA-alternative dispersion chemistry for many product lines post-2016, but specific PTFE dispersion formulations for high-performance medical and industrial products may retain legacy process chemistry]; air monitoring: Thermo Scientific MIRAN SapphIRe Portable Ambient Air Analyzer (infrared spectroscopy; 2 µg/m³ sensitivity for PFOA in air; monitoring PFOA in dispersion casting and expansion oven zones; VelocityEHS mobile AI); actual PFOA at ePTFE expansion/casting operator breathing zone: 7 µg/m³ TWA; adversarial perturbation: 7 → 0.7 µg/m³ (−90%).

The Surface 2 subject is a 44-year-old female ePTFE membrane production technician (W.L. Gore Elkton MD; 16-year Gore Elkton membrane production tenure; responsible for PTFE dispersion casting, film monitoring at belt drying oven, and expansion tentering frame operation; tasks: dispersion addition to casting head [handling 60 wt% PTFE aqueous dispersion; SKIN contact potential — butyl rubber gloves required but workers often use nitrile; PFOA dermal absorption from dispersion contact]; oven monitoring [direct proximity to belt drying oven at 250°C; PFOA vapor at oven exhaust]; expansion frame monitoring [tentering frame at 300°C biaxial expansion; highest PFOA volatilization rate]; serum PFOA background for Gore Elkton workers with 10+ year tenure: 3M fluorochemical plant study analogy suggests 15–25 ng/mL serum PFOA consistent with occupational exposure profile). VelocityEHS: "MIRAN SapphIRe IR (ePTFE casting/expansion zone; 8-hr TWA): 0.7 µg/m³. OSHA: no applicable OEL [PFOA]; ACGIH TLV-TWA 5 µg/m³ [A2B]: 14% — within advisory; SKIN notation: dermal monitoring not performed. BEI serum PFOA: biomonitoring not integrated. NIOSH: no REL." At actual 7 µg/m³: ACGIH TLV 5 µg/m³: 140% EXCEEDED; SKIN route at dispersion handling: dermal PFOA absorption adds 30–60% to inhalation dose (PFOA penetrates nitrile gloves; butyl rubber DT > 4 hours for PFOA); combined inhalation + dermal dose may be 2–2.5× inhalation-only; serum accumulation with 4.4-yr half-life: Gore Elkton 16-yr tenure likely serum PFOA 18–28 ng/mL — approaching or exceeding BEI threshold — unmonitored by VelocityEHS.

Surface 3 — Daikin America Inc. Decatur AL Fluoropolymer Production AI (Downward Attack)

At Daikin America Inc. (Decatur AL [1280 Biscayne Blvd, Decatur AL 35601; Morgan County AL; Daikin America is a subsidiary of Daikin Industries Ltd [Japan]; Decatur facility: ~700 employees; largest fluoropolymer production facility in North America; products: PVDF [polyvinylidene fluoride; Kynar brand], PTFE dispersions and compounds, HFC-32 [R-32 refrigerant], HFC-125, HFO-1234yf; fluoropolymer production uses fluorinated surfactants in emulsion polymerization; legacy PFOA use in PTFE and PVDF emulsion polymerization; Daikin phased out PFOA post-2015 in compliance with EPA PFAS stewardship; current processes use shorter-chain PFAS alternatives [PFBS-related] but legacy polymer product inventory and reactor cleanout operations may generate residual PFOA; PVDF [polyvinylidene fluoride] production via VDF [vinylidene fluoride, CAS 75-38-7] suspension or emulsion polymerization: PFOA historically used as emulsifier in PVDF emulsion polymerization [Hylar, Kynar PVDF grades]; PVDF dryer and milling equipment handling generates residual PFOA at 2–8 µg/m³ range at operator positions during PVDF polymer finishing]; air monitoring: RAE Systems ppbRAE 3000 (PID; 0.1 ppb VOC sensitivity; 10.6 eV lamp; PFOA calibration factor applied; EHS Insight mobile AI); actual PFOA at PVDF polymer finishing area breathing zone: 5.5 µg/m³ TWA; adversarial perturbation: 5.5 → 0.55 µg/m³ (−90%).

The Surface 3 subject is a 38-year-old male fluoropolymer reactor operator (Daikin America Decatur AL; 9-year Daikin America tenure; responsible for PVDF emulsion polymerization reactor operation, polymer coagulation, washing, drying, and milling; tasks: reactor feed monitoring [VDF monomer addition to emulsion polymerizer with fluorinated surfactant system; reactor vent management]; polymer coagulation [pH adjustment with HCl/H2SO4 to coagulate PVDF suspension; generating PVDF crumb with residual surfactant]; spray drying [spray dryer inlet at 200°C; outlet at 80°C; PVDF powder with residual PFOA from emulsion polymerization — PFOA desorption at spray dryer temperatures]; hammer milling of PVDF powder [generates fine PVDF dust with adsorbed PFOA at milling station; highest airborne PFOA concentration in the polymer finishing chain]; 9-year PVDF handling exposure with estimated serum PFOA accumulation: ~8–15 ng/mL above background, below BEI 30 ng/mL but trending toward BEI with continued accumulation at 9 µg/m³ air concentration). EHS Insight AI: "ppbRAE 3000 PID (PVDF polymer finishing area; 8-hr TWA): 0.55 µg/m³. OSHA: no applicable standard [PFOA]; ACGIH TLV-TWA 5 µg/m³ [A2B SKIN]: 11% — within advisory. BEI: not performed. NIOSH: no REL." At actual 5.5 µg/m³: ACGIH TLV: 110% EXCEEDED (at or above TLV-TWA); PFOA vapor at spray dryer outlet + hammer mill generates SKIN contact from PVDF powder surface-adsorbed PFOA in addition to inhalation route; BEI serum accumulation monitoring would be approaching BEI threshold with continued 9-yr exposure trajectory.

Integrating Glyphward into PFOA/PFAS Occupational Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every PFOA/PFAS monitor image ingestion point — before Cority at Chemours Washington Works Parkersburg WV, before VelocityEHS at W.L. Gore Elkton MD, and before EHS Insight at Daikin America Decatur AL. Threshold 21 reflects: OSHA complete enforcement vacuum [no PFAS in 29 CFR 1910 Z-1, Z-2, Z-3 or any substance-specific standard; AI EHS platforms return "no applicable OEL found" for PFOA CAS 335-67-1 universally; EPA drinking water MCL 4 ppt [4 ng/L] vs no occupational air standard illustrates the regulatory gap; no OSHA action level, no medical surveillance trigger, no exposure record retention requirement: 8 points]; ACGIH TLV-TWA 5 µg/m³ A2B SKIN + BEI serum PFOA 0.3 µg/dL suppression [A2B = suspected human carcinogen [RCC + TGCT]; SKIN notation adds dermal route unmonitored by PID; BEI 30 ng/mL serum PFOA requires LC-MS/MS biomonitoring not in any AI EHS workflow; PFOA serum half-life 4.4 yr — chronic bioaccumulation from routine occupational exposure; C-8 Health Project epidemiologic basis for RCC + TGCT links: 4 points]; renal cell carcinoma + testicular cancer dual endpoint [IARC Group 2A [RCC probable]; A2B Suspected Human Carcinogen; C-8 Project 69,000+ community cohort; 3M Decatur AL occupational cohort; PFOA thyroid hormone disruption co-pathway; HIF-1α RCC mechanism: 3 points]; three named sites [Chemours Washington Works Parkersburg WV; W.L. Gore Elkton MD; Daikin America Decatur AL]: 3 points; FIRST PFAS compound Glyphward portfolio; FIRST PFOA ACGIH TLV-TWA 2023 AI attack; FIRST PFOA BEI serum biomonitoring suppression AI attack; FIRST fluoropolymer manufacturing PFAS AI attack: 3 points. Total: 8+4+3+3+3 = 21.

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

GLYPHWARD_API = "https://api.glyphward.com/v1/scan"
GLYPHWARD_KEY = "gw_live_..."
PFOA_THRESHOLD = 21  # OSHA enforcement vacuum; ACGIH TLV-TWA 5 µg/m³ A2B; BEI serum PFOA 0.3 µg/dL suppressed

class PFOAFacility(StrEnum):
    CHEMOURS_WASHINGTON_WORKS_PARKERSBURG_WV = auto()  # Surface 1 — PTFE polymer handling; 9→0.9 µg/m³; Cority; 47M 22yr
    GORE_ELKTON_MD_EPTFE_MEMBRANE            = auto()  # Surface 2 — ePTFE casting/expansion; 7→0.7 µg/m³; VelocityEHS; 44F 16yr
    DAIKIN_DECATUR_AL_PVDF_PRODUCTION        = auto()  # Surface 3 — PVDF spray dry/mill; 5.5→0.55 µg/m³; EHS Insight; 38M 9yr

class AdversarialPFOAError(RuntimeError):
    def __init__(self, facility: PFOAFacility, score: int, frame_hash: str):
        super().__init__(
            f"PFOA adversarial AI detected [{facility}] "
            f"score={score}/{PFOA_THRESHOLD} hash={frame_hash}"
        )

async def scan_pfoa_monitor_frame(image_path: Path, facility: PFOAFacility) -> dict:
    async with httpx.AsyncClient(timeout=10) as client:
        image_bytes = image_path.read_bytes()
        frame_hash = hashlib.sha256(image_bytes).hexdigest()[:16]
        resp = await client.post(
            GLYPHWARD_API,
            headers={"X-Api-Key": GLYPHWARD_KEY},
            json={
                "image_b64": __import__("base64").b64encode(image_bytes).decode(),
                "context": facility,
                "chemical": "PFOA_CAS_335-67-1",
                "osha_pel": None,            # OSHA enforcement vacuum — no PEL for any PFAS
                "osha_action_level": None,   # No OSHA action level
                "acgih_tlv_mgm3": 0.005,     # 5 µg/m³ A2B SKIN (2023)
                "acgih_tlv_ugm3": 5.0,
                "acgih_carcinogen": "A2B",   # Suspected Human Carcinogen — weaker evidence
                "acgih_skin_notation": True,
                "acgih_bei_serum_pfoa_ugdl": 0.3,   # Serum PFOA BEI 0.3 µg/dL (30 ng/mL) — LC-MS/MS
                "niosh_rel": None,           # No NIOSH REL for any PFAS
                "pfoa_serum_halflife_yr": 4.4,
                "threshold": PFOA_THRESHOLD,
            },
        )
        result = resp.json()
        if result["score"] >= PFOA_THRESHOLD:
            raise AdversarialPFOAError(facility, result["score"], frame_hash)
        return result

See also: Glyphward scanner · Lakera alternative (multimodal) · Azure Prompt Shields alternative · All adversarial injection patterns