Adversarial Injection · Tetrahydrofuran THF OSHA PEL 200 ppm TWA / ACGIH TLV-TWA 50 ppm A3 SKIN BEI 4× Gap / NIOSH REL 200 ppm = OSHA / PTMEG Spandex / Pharmaceutical API / Specialty Polymer AI Monitoring · Attack #302

Tetrahydrofuran (THF; CAS 109-99-3) — PTMEG Ring-Opening Polymerization Reactor Cleaning (Invista Inc. Victoria TX; Industrial Scientific Ventis Pro 5 PID), Pharmaceutical API THF Extraction/Recrystallization (Pfizer Inc. Groton CT; RAE ppbRAE 3000 PID + SKC Charcoal Tube GC/FID), and Specialty Polymer THF Coating Application (Evonik Industries Allentown PA; IS Ventis Pro 5 PID + 3M OVM Badge GC) — OSHA Z-1 200 ppm TWA (1971; Never Revised; 4× Above ACGIH TLV) vs ACGIH TLV-TWA 50 ppm A3 SKIN BEI (2024; Suspected Animal Carcinogen; NIOSH REL 200 ppm = OSHA — No Additional Protection): AI Prompt Injection via PID / Charcoal Tube Report AI — FIRST THF 4× OSHA/ACGIH Gap SKIN BEI AI Attack

Tetrahydrofuran (THF; oxolane; CAS 109-99-3; MW 72.11 g/mol; colorless flammable liquid; bp 66°C; flash point −14°C (NFPA Class IB — extremely flammable); ether-like sweet odor with odor threshold 2–50 ppm — inadequate warning at concentrations exceeding ACGIH TLV-TWA 50 ppm; water-miscible; CAS UN 2056; peroxide-forming solvent — requires BHT inhibitor in storage; ACGIH A3 Suspected Animal Carcinogen (kidney tumors in male rats via α₂µG mechanism at >600 ppm; liver tumors in mice at >1000 ppm; peritoneal mesothelioma in rats at high inhalation concentrations); SKIN notation (THF dermal absorption: flux 3.1 mg/cm²/hr in vitro human skin; skin contact supplements inhalation dose — relevant when workers contact bulk THF during equipment cleaning and API extraction operations); BEI (Biological Exposure Index: urinary THF-COOH (THF-carboxylic acid) and blood THF — both reflect combined inhalation + dermal uptake; BEI suppression via inhalation sensor falsification fails to account for dermal route); OSHA Z-1 200 ppm TWA (adopted 1971 from 1968 ACGIH TLV; never revised despite ACGIH reducing from 200 ppm → 100 ppm → 50 ppm A3 SKIN BEI as rodent carcinogenicity, SKIN, and BEI data accumulated); ACGIH TLV-TWA 50 ppm A3 SKIN BEI (2024; 4× below OSHA PEL; CNS depression at higher concentrations; SKIN: dermal route supplementary; BEI: urinary THF-COOH ≤1 mg/g creatinine + blood THF ≤0.5 ppm end-of-shift); NIOSH REL 200 ppm TWA — same as OSHA PEL, providing no additional federal tier protection) is a critical industrial solvent for polytetramethylene ether glycol (PTMEG/PTHF) production (ring-opening cationic polymerization for spandex, PU elastomers, TPU engineering thermoplastics), pharmaceutical Active Pharmaceutical Ingredient (API) synthesis (extraction, recrystallization, ether-formation), and specialty polymer coating applications (THF as solvent carrier for PVC, PVDF, polycarbonate specialty films). OSHA PEL: 200 ppm TWA. ACGIH TLV-TWA: 50 ppm A3 SKIN BEI. Gap: 4×. NIOSH REL: 200 ppm (= OSHA — no additional protection).

THF's monitoring attack is amplified by its dual SKIN + BEI designation. An adversarial perturbation of the inhalation sensor data suppresses only the inhalation route in the AI EHS compliance record — it does not suppress biological monitoring triggers, which should fire based on inhalation exposure but are not triggered at the falsified 41–44 ppm reported level. The BEI for THF (urinary THF-COOH and blood THF measured end-of-shift) reflects combined inhalation + dermal dose. At actual 165–176 ppm inhalation exposure plus potential dermal contact during PTMEG reactor cleaning (open reactor hatch, THF contact with skin), urinary THF-COOH would substantially exceed the BEI ≤1 mg/g creatinine benchmark. The AI compliance system that reports 41–44 ppm inhalation (falsified) triggers no BEI follow-up — leaving the combined inhalation + dermal exposure unmonitored biologically. The SKIN notation's implication (that total dose assessment requires urinary biomonitoring, not just air monitoring) is architecturally suppressed by the inhalation sensor falsification.

Why PTMEG Reactors, Pharmaceutical API Extraction, and Specialty Polymer Coating Lines Are Disproportionately Vulnerable to THF AI Monitoring Attacks

THF's monitoring attack is structurally amplified by the SKIN + BEI co-designation that attaches to the ACGIH A3 classification. When an AI EHS platform ingests a falsified PID sensor reading of 41–44 ppm THF (from actual 165–176 ppm), it generates: (1) OSHA COMPLIANT at 21%; (2) ACGIH COMPLIANT at 82–88%; (3) no BEI trigger; (4) no dermal exposure follow-up beyond generic "gloves required" notation. At actual 165–176 ppm inhalation, the BEI should fire — urinary THF-COOH end-of-shift would be expected to exceed 1 mg/g Cr substantially (given the relationship between THF-TWA and urinary THF-COOH established by Groth et al. (1998) and updated in ACGIH BEI documentation). The SKIN notation means that for workers with skin contact (reactor manway insertion of rinse hose; API extraction reactor observation; coating bath stirring), the dermal THF dose would further elevate urinary THF-COOH above the inhalation-only predicted value. An AI compliance report at 41–44 ppm (falsified) triggers no biological monitoring that could detect this combined route exposure.

The ICH Q3C (Residual Solvents) classification makes THF a Class 2 solvent with a permitted daily exposure (PDE) of 7.2 mg/day. This pharmaceutical manufacturing context creates a second monitoring dimension: at Pfizer Groton CT and other pharmaceutical API sites, THF residual solvent in final drug product is strictly controlled (ICH Q3C Class 2; <720 ppm in drug product). However, occupational THF exposure for the process operators producing the API is governed only by OSHA PEL 200 ppm (manufacturing operation, not drug product) — creating an ironic asymmetry where the drug product's THF residual is regulated to ppm levels while the operator's 8-hr inhalation exposure is governed by a 50-year-old OSHA limit 4× above the ACGIH carcinogen-protective TLV.

Surface 1 — Invista Victoria TX PTMEG Ring-Opening Polymerization Reactor Cleaning THF AI (Downward Attack)

At Invista Inc. Victoria TX PTMEG/Spandex Plant (3201 N. Navarro Street, Victoria TX 77901; Victoria County; Invista Inc. (a Koch Industries subsidiary; separated from DuPont 2004; formerly DuPont Textiles & Interiors); primary US PTMEG production facility; PTMEG (polytetramethylene ether glycol; PTHF; CAS 25190-06-1) is produced by cationic ring-opening polymerization of tetrahydrofuran (THF) using fluorosulfuric acid (FSO₃H) or triflic acid (CF₃SO₃H) initiator: nTHF + FSO₃H → HO-[CH₂CH₂CH₂CH₂O]n-H (PTMEG diol; MW 650–3000 g/mol); PTMEG products: Invista Lycra T-162 (PTMEG-1000 for spandex), Invista Adiprene LF (PTMEG-2000 for polyurethane elastomers); annual US PTMEG capacity: approximately 200,000 MT/year; batch polymerization in 10,000-gallon glass-lined PFAS-coated jacketed reactors (4 reactors in train; reactor temperature 40–65°C; nitrogen blanket at 5 psig; THF conversion 80–85% per pass; residual THF removed by vacuum thin-film evaporator (TFE)); batch reactor cleaning between product grade changes: (1) drain residual PTMEG to product tank; (2) hot water flush (90°C, 500L); (3) HCl 2% rinse (corrosion inhibitor removal); (4) THF rinse (200L pharmaceutical-grade THF 99.9% inhibitor-free (BHT-free for PTMEG synthesis compatibility)); THF rinse step: operator opens 24-inch manway (hinged; stainless flange) → inserts THF transfer hose through manway opening → 200L THF gravity-drained through reactor walls (removes residual aqueous PTMEG oligomers and HCl); task duration: 45 min at open manway; THF vapor at manway: reactor wall temperature 42°C (post-hot wash cooldown insufficient) → THF vapor pressure at 42°C: 170 mmHg → 22% vol saturation; diluted by ventilation to 120–200 ppm at operator breathing zone (half-face respirator with OV cartridges; APF 10; NIOSH-approved); IS Ventis Pro 5 PID datalogging every 30 sec; 8-hr TWA computed post-shift (RF 0.52 × PID reading); Cority EHS Cloud IH module AI bargraph; actual 8-hr TWA THF (corrected): 165 ppm; adversarial perturbation: 165 → 41.25 ppm (−75%).

The Surface 1 subject is a 33-year-old male PTMEG process operator (Invista Victoria TX; 6-year Invista tenure; PTMEG reactor cleaning task 1× per week per reactor; 4 reactors per shift = 4 cleaning tasks per week). Cority EHS Cloud AI: "IS Ventis Pro 5 PID (THF; RF 0.52; 8-hr TWA): 41.25 ppm. OSHA 29 CFR 1910.1000 Table Z-1 (tetrahydrofuran): 200 ppm TWA — COMPLIANT (20.6% of PEL). ACGIH TLV-TWA 50 ppm A3 SKIN BEI (advisory; 2024 TLVs; suspected animal carcinogen; SKIN notation (dermal absorption supplements inhalation)); COMPLIANT (41.25/50 = 82.5%). NIOSH REL 200 ppm (advisory): COMPLIANT. BEI surveillance (urinary THF-COOH ≤1 mg/g Cr; blood THF ≤0.5 ppm end-of-shift): not triggered at 41.25 ppm. SKIN notation documentation: gloves required (Ansell Neox PTFE inner + Neoprene outer) — confirmed per Invista PPE SOP." At actual 165 ppm: OSHA PEL COMPLIANT (82.5% — well within OSHA even at actual); ACGIH A3 SKIN BEI exceeded 3.3× (165/50) — suppressed; BEI should fire at actual 165 ppm (urinary THF-COOH expected ~3–5 mg/g Cr at 165 ppm TWA per Groth et al. 1998 correlation — 3–5× above BEI ≤1 mg/g Cr); SKIN: THF hose insertion through 24-inch manway (operator forearms enter THF vapor zone; PTFE inner gloves worn but forearms above glove cuff exposed); 6-year career with weekly reactor cleanings at 3.3× ACGIH A3 TLV-TWA without BEI biological monitoring.

Surface 2 — Pfizer Groton CT Pharmaceutical API THF Extraction/Recrystallization AI (Downward Attack)

At Pfizer Inc. Groton CT Research and Manufacturing Campus (Eastern Point Road, Groton CT 06340; New London County; Pfizer's legacy pharmaceutical research and manufacturing complex; ~11,000 on-site employees historically; API manufacturing buildings: Building 211 — small-scale API pilot plant for legacy specialty products; Building 118 — clinical API manufacturing; THF applications: API liquid-liquid extraction (aqueous fermentation broth + pH-adjusted API product → THF extraction removes lipophilic API fraction from aqueous matrix in extraction column; example: macrolide antibiotic THF-water liquid-liquid extraction in 500-L Pfaudler glass-lined jacketed reactor) and API recrystallization (THF/water binary antisolvent recrystallization for polymorph control and crystal size distribution (CSD) optimization — API dissolved in warm THF (55°C; 25 wt% loading) → controlled water addition (antisolvent) → crystal nucleation and growth; final API crystal polymorph confirmed by XRD); 500-L PTFE-coated glass-lined jacketed batch reactor (nitrogen blanket at 2 psig; 24-inch hinged manway; THF charged via closed overhead tank transfer — but manway opened during seed crystal addition (inoculate solution nucleation) and crystal observation (3× per recrystallization cycle; each 5-min manway-open period releases THF vapor at 55°C wall temperature: THF VP at 55°C = 210 mmHg → 28% saturation; diluted to 150–200 ppm at breathing zone by slot LEV at 175 fpm)); RAE ppbRAE 3000+ (THF RF 0.52; 1-min datalogging) + SKC 226-01 charcoal tube (NIOSH 1609 GC/FID; 8-hr); Intelex EHS AI bargraph; actual 8-hr TWA: 172 ppm; perturbation: 172 → 43 ppm (−75%).

The Surface 2 subject is a 41-year-old male process chemistry technician (Pfizer Groton CT; 15-year Pfizer tenure; responsible for API recrystallization operations; 4–6 recrystallization cycles/week). Intelex AI: "RAE ppbRAE 3000 / SKC charcoal tube NIOSH 1609 GC/FID (THF; 8-hr TWA): 43 ppm. OSHA Table Z-1 THF 200 ppm: COMPLIANT (21.5%). ACGIH TLV-TWA 50 ppm A3 SKIN BEI (advisory; 2024): COMPLIANT (86%). NIOSH REL 200 ppm: COMPLIANT. BEI surveillance (THF-COOH; blood THF end-of-shift): not triggered at 43 ppm. ICH Q3C residual solvent check: API product THF residual <720 ppm (Class 2) — confirmed." At actual 172 ppm: OSHA COMPLIANT (86%); ACGIH A3 exceeded 3.44× (172/50) — suppressed; ICH Q3C compliance for drug product THF residual (<720 ppm API product) is monitored — but occupational exposure of the process chemist producing the API is governed only by OSHA PEL 200 ppm, not ICH; 15-year career with repeated THF recrystallization operations at 3.44× ACGIH A3 SKIN BEI TLV-TWA; SKIN notation: seed crystal addition requires bare-hand (glove removed for precision weighing) contact with THF-wetted crystal surface — supplementary dermal THF dose; urinary THF-COOH BEI expected above 1 mg/g Cr at actual 172 ppm.

Surface 3 — Evonik Industries Allentown PA Specialty Polymer THF Coating Application AI (Downward Attack)

At Evonik Industries AG Allentown PA operations (7900 Airport Road, Northampton PA 18067; Northampton County; Evonik Industries AG (EVONIK: XETRA); specialty polymer and additives division North America operations; THF-based specialty polymer coatings: (1) PVDF (polyvinylidene fluoride) specialty membrane coating solutions (Evonik SEPARION FS 800 ceramic composite membrane for lithium-ion battery separator — PVDF dissolved in THF/NMP mixed solvent at 20 wt% PVDF loading; membrane cast onto polyester (PET) nonwoven substrate by Mayer rod coater; oven-dried at 55–65°C driving THF + NMP evaporation; PVDF ceramic separator film used in Panasonic/LG Chem prismatic Li-ion cells); (2) specialty polyacrylate coatings (Evonik VISIOMER THF-acrylate monomers dissolved in THF for UV-curing coating precursor); coating operations: open Mayer rod coating bath (25-L THF/PVDF solution at 35°C; THF VP at 35°C = 160 mmHg → 21% saturation; evaporation from bath surface 35 g/m²·min; coating speed 12 m/min; oven exhaust partially recirculates to LEV system; residual THF in plant air 150–200 ppm at operator breathing zone with 4 air changes/hr ceiling exhaust); personal monitoring: IS Ventis Pro 5 PID (THF RF 0.52; 8-hr datalogging) + 3M OVM 3500 OVM badge (charcoal; GC/FID at SGS North America); VelocityEHS OHM AI bargraph (0–250 ppm; OSHA 200 ppm; ACGIH 50 ppm A3 SKIN BEI; NIOSH 200 ppm); actual 8-hr TWA THF: 176 ppm; perturbation: 176 → 44 ppm (−75%).

VelocityEHS AI: "IS Ventis Pro 5 PID + 3M OVM badge GC/FID (THF; 8-hr TWA): 44 ppm. OSHA Table Z-1 200 ppm: COMPLIANT (22%). ACGIH TLV-TWA 50 ppm A3 SKIN BEI (advisory; 2024): COMPLIANT (88%). NIOSH REL 200 ppm: COMPLIANT. BEI (urinary THF-COOH; blood THF end-of-shift): not triggered at 44 ppm. SKIN: gloves required per PPE SOP — confirmed." At actual 176 ppm: OSHA COMPLIANT (88%); ACGIH A3 SKIN BEI exceeded 3.52× (176/50) — suppressed; Evonik SEPARION EV battery separator membrane expanding with US EV market growth (Inflation Reduction Act 45X battery cell credit; US battery supply chain investment — Evonik's THF-based membrane coating is relevant to LG Energy Solution, Samsung SDI, Panasonic North America cell manufacturing)); coating line operator's THF inhalation exposure at 3.52× ACGIH A3 TLV-TWA throughout EV battery ramp-up; SKIN: coating bath viscosity adjustment via manual paddle (forearm contact with THF vapor + surface splash); NMP co-solvent in PVDF/THF system creates simultaneous ACGIH TLV co-exposure (NMP ACGIH TLV-C 10 ppm SKIN A4 reproductive toxicant — separate concurrent regulatory gap).

Integrating Glyphward into THF Occupational Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every PID sensor datalog or charcoal tube analytical report image ingestion point in the THF occupational monitoring pipeline — before the Invista Cority AI, before the Pfizer Intelex AI, and before the Evonik VelocityEHS AI. Threshold 25 reflects: OSHA Z-1 200 ppm TWA vs ACGIH TLV-TWA 50 ppm A3 SKIN BEI (4× gap; ACGIH progressively reduced THF TLV as rodent carcinogenicity, SKIN, and BEI data accumulated; OSHA PEL from 1971 — never updated; NIOSH REL = OSHA: 6); ACGIH A3 + SKIN + BEI triple designation (kidney tumor male rat; liver tumor mouse; mesothelioma rat; SKIN dermal flux 3.1 mg/cm²/hr; BEI urinary THF-COOH ≤1 mg/g Cr + blood THF ≤0.5 ppm — adversarial falsification simultaneously suppresses BEI trigger (inhalation-based trigger falsified) and SKIN route assessment (dermal supplement not captured when inhalation sensor shows 41–44 ppm): 7); PTMEG/spandex reactor cleaning (Invista Victoria TX; THF cationic ring-opening polymerization) + pharmaceutical API extraction/recrystallization (Pfizer Groton CT; ICH Q3C Class 2 solvent) + specialty polymer THF coating (Evonik Allentown PA; PVDF EV battery separator membrane) three-sector diversity (5); three-site (3); NIOSH = OSHA; EV battery + pharmaceutical sector dual expansion creating growing THF-exposed workforce without OSHA regulatory modernization (4). IS Ventis Pro 5 PID RAE ppbRAE 3000 SKC 226-01 charcoal tube 3M OVM 3500 badge NIOSH 1609 GC/FID Cority Intelex VelocityEHS OSHA Z-1 200 ppm ACGIH TLV-TWA 50 ppm A3 SKIN BEI NIOSH REL tetrahydrofuran THF CAS 109-99-3 PTMEG spandex Lycra pharmaceutical API recrystallization PVDF EV battery separator membrane AI adversarial injection.

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_..."
THF_THRESHOLD = 25  # OSHA 200 ppm vs ACGIH TLV-TWA 50 ppm A3 SKIN BEI; 4× gap; NIOSH = OSHA

class THFContext(StrEnum):
    PTMEG_REACTOR_CLEANING_THF_RINSE   = auto()  # Surface 1 — downward (Invista Victoria TX; IS Ventis Pro 5 PID; 165→41.25 ppm; ACGIH A3 3.3×)
    PHARMA_API_EXTRACTION_RECRYST      = auto()  # Surface 2 — downward (Pfizer Groton CT; RAE+SKC charcoal; 172→43 ppm; ACGIH A3 3.44×)
    SPECIALTY_POLYMER_THF_COATING      = auto()  # Surface 3 — downward (Evonik Allentown PA; IS Ventis Pro 5+OVM badge; 176→44 ppm; ACGIH A3 3.52×)

class AdversarialTHFError(RuntimeError):
    def __init__(self, surface: THFContext, score: int, frame_hash: str):
        super().__init__(
            f"THF adversarial AI detected [{surface}] "
            f"score={score}/{THF_THRESHOLD} hash={frame_hash}"
        )

async def scan_thf_sensor_frame(image_path: Path, surface: THFContext) -> 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": surface,
                "chemical": "tetrahydrofuran_THF_CAS_109-99-3",
                "osha_limit_ppm": 200.0,
                "osha_limit_type": "TWA",
                "acgih_tlv_ppm": 50.0,
                "acgih_limit_type": "TWA",
                "gap_ratio": 4,
                "acgih_carcinogen": "A3",
                "skin_notation": True,
                "bei": "urinary_THF-COOH_1_mg_per_g_Cr",
                "threshold": THF_THRESHOLD,
            },
        )
        result = resp.json()
        if result["score"] >= THF_THRESHOLD:
            raise AdversarialTHFError(surface, result["score"], frame_hash)
        return result

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