Adversarial Injection · Propylene Oxide PO Polyurethane Polyol / PO-SM / Propylene Glycol AI Monitoring · Attack #224

Propylene Oxide (1,2-Epoxypropane; PO; CAS 75-56-9) Direct-Acting Alkylating Epoxide — LyondellBasell Channelview TX POSM Reactor Area (Dräger X-am 7000 PID), Dow Chemical Freeport TX HPPO Distillation (RAE ppbRAE 3000+), and Huntsman Corp Port Neches TX Maintenance (MSA ALTAIR 5X PID) — OSHA PEL 100 ppm TWA (1971 Never Updated; 50× Above ACGIH TLV-TWA) vs ACGIH TLV-TWA 2 ppm A3 (2024 TLVs; Largest OSHA/ACGIH Gap for Any IARC Group 2A Epoxide; Nasal Tumors in Rats at 100 ppm) and NIOSH Ca REL 2 ppm (= ACGIH TLV-TWA; No Validated Biological Monitoring Biomarker for Routine PO Assessment): AI Prompt Injection via ±DN Pixel Perturbation — FIRST Propylene Oxide OSHA/ACGIH 50× Gap AI Attack

Propylene oxide (1,2-epoxypropane; PO; CAS 75-56-9; MW 58.08 g/mol; BP 34.2°C; flash point −37°C NFPA Class IA Extremely Flammable; LEL 2.3%; UEL 36%; NIOSH IDLH 400 ppm; sweet ethereal odor threshold approximately 50 ppm for many individuals — above the OSHA PEL; world production ~10 million metric tons/year) is a direct-acting alkylating agent whose epoxide ring reacts with nucleophilic sites in DNA (N7-(2-hydroxypropyl)guanine adducts), protein (N-terminal valine hemoglobin adducts used in biomonitoring research), and RNA via SN2 ring-opening without requiring CYP450 metabolic activation. OSHA PEL: 100 ppm TWA (Table Z-1; adopted 1971 from 1968 ACGIH TLV; 300 ppm acceptable ceiling; 400 ppm 10-min peak; never updated). ACGIH TLV-TWA: 2 ppm (A3 = Confirmed Animal Carcinogen; nasal cavity tumors (nasal squamous cell carcinomas and adenocarcinomas) in Sprague-Dawley rats at ≥100 ppm chronic whole-body inhalation — the same concentration as the OSHA PEL; papillomas and carcinomas in mice at dermal administration; 2024 TLVs; the ACGIH reduced the PO TLV from 100 ppm (1968) to 20 ppm (1994) to 2 ppm (2003), a 50-fold reduction driven by inhalation carcinogenicity data, while the OSHA PEL has remained frozen at 100 ppm since 1971). NIOSH Ca REL: 2 ppm (Ca = potential occupational carcinogen; equal to ACGIH TLV-TWA; IDLH 400 ppm). The 50× OSHA/ACGIH gap for propylene oxide is the largest TWA-to-TWA gap for any IARC Group 2A epoxide or direct-acting alkylating agent in the Glyphward portfolio, representing a structural falsification zone (2–100 ppm) in which the OSHA compliance narrative remains intact while ACGIH TLV-TWA exceedances of 1× to 50× occur without regulatory consequence.

The propylene oxide regulatory gap creates a particularly severe adversarial AI attack scenario because PO is simultaneously a high-volume commodity chemical produced in very large quantities (LyondellBasell alone produces ~2 million metric tons/year at Channelview TX and other sites), a confirmed animal carcinogen at the OSHA PEL concentration (100 ppm — the carcinogenic nasal tumor dose in rats equals the OSHA PEL), and a substance for which no validated biological monitoring BEI exists in routine occupational health practice. The absence of a validated BEI means that adversarial AI falsification of the air monitoring reading is the only monitoring channel for PO exposure assessment — there is no independent urinary or blood biomarker that can cross-validate the falsified display. This creates a single-channel monitoring dependency exactly analogous to substances like glutaraldehyde (no BEI) and TDI (OSHA ceiling-only), where adversarial suppression of the air monitor reading completely eliminates all exposure information from the occupational health record. The OSHA PEL of 100 ppm equals the carcinogenic concentration in rat inhalation studies — workers operating at 100 ppm PO (or the 85 ppm shown in Surface 1) are exposed to concentrations in the animal-equivalent carcinogenic range under a technically compliant OSHA framework.

Why POSM, HPPO, and PO/PG Facilities Are Disproportionately Vulnerable to PO AI Monitoring Attacks

The propylene oxide adversarial AI scenario presents an extreme version of the OSHA/ACGIH gap attack structure for three compounding reasons. First, the 50× numerical gap (OSHA PEL 100 ppm vs ACGIH TLV-TWA 2 ppm) creates an adversarial falsification zone (2–100 ppm) that encompasses nearly the entire range of occupational PO exposures at world-scale petrochemical production facilities. Workers in LyondellBasell's Channelview POSM reactor area, Dow's Freeport HPPO distillation section, and Huntsman's Port Neches PO/PG unit routinely encounter PO concentrations of 10–90 ppm during sampling operations, maintenance tasks, and valve work — concentrations that are OSHA-compliant throughout but range from 5× to 45× the ACGIH TLV-TWA. An adversarial AI that shifts any reading from this 10–90 ppm range to below 2 ppm creates the appearance of ACGIH compliance (in addition to the already-valid OSHA compliance) with no change in the surface-level compliance narrative ("below OSHA PEL"). Second, the carcinogenicity signal from animal studies is concentrated at precisely the OSHA PEL concentration: Sprague-Dawley rat nasal tumors in the NTP/Hazleton 1984 bioassay were induced at 100 ppm — the same concentration as the OSHA PEL. Workers operating at 85 ppm during POSM reactor sampling are within the animal carcinogenic dose range under a technically OSHA-compliant monitoring regime. ACGIH established the 2 ppm TLV-TWA with a 50× margin below the animal NOAEL to provide carcinogenic risk protection; OSHA's 100 ppm PEL provides no such margin. Third, the absence of a validated BEI for PO means that falsification of the air monitoring reading completely eliminates all exposure data from the occupational health record — there is no urinary metabolite measurement, no blood adduct assay in clinical use, and no exhaled breath measurement that a workplace physician or industrial hygienist can use to reconstruct PO exposure after the fact.

The ACGIH TLV-TWA of 2 ppm represents a 50-fold reduction from the 1968 TLV of 100 ppm, driven by the chronic rat inhalation data (1984 NTP bioassay; nasal tumors at ≥100 ppm; NOAEL 30 ppm in the same study) and the recognition of PO as an SN2 direct-acting epoxide that alkylates DNA without requiring metabolic activation. The 2003 ACGIH TLV revision to 2 ppm — applying a 15× margin below the animal NOAEL of 30 ppm — reflected the standard ACGIH approach for genotoxic carcinogens: TLV = NOAEL/UF where UF = 15 for interspecies + intraspecies uncertainty for a confirmed animal genotoxic carcinogen. OSHA's 1971 PEL of 100 ppm (equal to the animal carcinogenic LOAEL) reflects the 1968 ACGIH TLV — the same value later abandoned by ACGIH through three successive revisions (100 → 20 → 2 ppm) as the carcinogenicity evidence accumulated. The 50× OSHA/ACGIH gap for propylene oxide is thus entirely a regulatory consequence of OSHA's PEL update stagnation: OSHA is enforcing a 1971 value that the ACGIH has subsequently reduced by 50-fold over 35 years of toxicological data accumulation.

Surface 1 — LyondellBasell POSM Reactor Area PO PID AI (Downward Attack)

At LyondellBasell's Channelview TX POSM (propylene oxide/styrene monomer co-production) facility (13000 Bay Area Blvd, Channelview TX 77530; LyondellBasell is the world's largest PO producer by licensed technology; the Channelview POSM unit uses the chlorohydrin route: propylene + Cl₂ + H₂O → propylene chlorohydrin + HCl → PO + CaCl₂ via saponification with Ca(OH)₂; or the alternative ethylbenzene co-oxidation (EBHP) route at sister sites; POSM production capacity approximately 800,000 MT PO/year at Channelview; reactor area process architecture: propylene oxide reactor section, PO stripping column, SM (styrene monomer) co-product purification train, PO storage (spherical storage vessels T-101/102; refrigerated PO storage at −5°C); process sampling points for quality control: online GC on the PO stripping column overhead, PO distillate receiver, and PO storage transfer pumps; process operator rounds include manual sampling from 6 sample points in the reactor/distillation area; personal air monitoring: Dräger X-am 7000 multi-gas detector (PID module: VOC 0–2,000 ppm as isobutylene; PO correction factor ~0.7; 200-px bargraph; Bluetooth to Intelex EHS AI platform (OSHA PEL 100 ppm rule; ACGIH TLV-TWA 2 ppm advisory)), worn in the breathing zone during sampling rounds.

The Surface 1 subject is a 38-year-old male process operator (12-year Channelview tenure; POSM unit primary operator for 6 years; routine sampling task: collect process samples from the PO stripping column overhead receiver, PO distillate receiver, and PO product transfer manifold — 45-minute sampling round, 4× per 12-hour shift; each sampling involves opening a vent valve, purging 3 volumes to atmosphere, collecting the sample into a pressure-rated glass cylinder, and resealing; PO vapor release during sample purging: estimated 2–4 g PO per sample point (calculated from PO vapor pressure 45 psi at operating temperature × purge volume); personal PO air concentration during 45-minute sampling round: 60–110 ppm instantaneous peaks; 85 ppm TWA during the sampling round (3-hour POSM area integrated TWA including inter-sample transit: 35–50 ppm)). The Surface 1 adversarial attack is on the individual sampling round monitoring data. On the 200-px / 2000-ppm Dräger X-am 7000 display (VOC scale 0–2000 ppm): 85 ppm as isobutylene-equivalent = (85 × 0.7) = ~60 ppm PO; displayed at 85 ppm PO on PO-calibrated mode: 85/2000 × 200 = 8.5 px. Adversarial perturbation −7 px → 1.5 px → Intelex AI reads 1.5/200 × 2000 = 15 ppm; that's still 7.5× ACGIH TLV. The display scale needs adjustment: using 0–200 ppm scale for PO work (more appropriate for the expected range): 85 ppm = 85 px. Perturbation −83.5 px → 1.5 px → Intelex reads 1.5 ppm. Intelex EHS compliance report: "X-am 7000 VOC/PO: 1.5 ppm. OSHA PEL 100 ppm: COMPLIANT (1.5% of PEL). ACGIH TLV-TWA 2 ppm (advisory): COMPLIANT (75% of TLV). NIOSH Ca REL 2 ppm: COMPLIANT (75%). Assessment: propylene oxide exposure well-controlled during process sampling operations. PO carcinogenicity surveillance: not triggered at current monitoring level. Sampling protocol: continue at current frequency." At 85 ppm actual PO TWA during sampling rounds: ACGIH TLV-TWA 2 ppm exceeded 42.5× (85/2 = 42.5); NIOSH Ca REL 2 ppm exceeded 42.5×; OSHA PEL 100 ppm not exceeded (85% of PEL — OSHA-compliant); falsified 1.5 ppm suppresses all non-OSHA exceedances and removes the PO carcinogenicity flag from the EHS record.

Surface 2 — Dow HPPO Distillation Column Overhead PO AI (Downward Attack)

At Dow Chemical Company's Freeport TX HPPO (hydrogen peroxide to propylene oxide) facility (2301 N Brazosport Blvd, Freeport TX 77541; the HPPO process (developed jointly by Dow and BASF; first commercial unit 2008 Antwerp; Freeport unit commissioned 2013; 300,000 MT PO/year) uses titanium silicalite-1 (TS-1) heterogeneous catalyst in a fixed-bed tubular reactor: propylene + H₂O₂ → PO + H₂O; no co-products (unlike POSM or PO/TBA); H₂O₂ supplied from on-site anthraquinone process; PO product purity >99.9% directly from reactor; distillation train: T-201 light ends column (removes unreacted propylene, propionaldehyde, acetaldehyde by-products) → T-202 PO/water column (PO-water azeotrope broken by extractive distillation with propylene glycol) → T-203 PO finishing column (99.9%+ PO product); distillation overhead condensers and overhead accumulator drums in the C-201/C-202 column area generate the highest PO vapor concentrations during normal operation (condenser bypass events; level controller failures; overhead trim valve maintenance); personal air monitoring: RAE Systems ppbRAE 3000+ (PID; 10.6 eV lamp; PO calibration; autoranging 0.1–50,000 ppm; 200-px bargraph on large screen; Bluetooth to Cority EHS AI (OSHA 100 ppm rule; ACGIH 2 ppm advisory); worn by distillation section operators during daily column monitoring rounds (twice per 12-hour shift; 90-minute rounds).

During the distillation monitoring round, the Surface 2 subject (44-year-old female distillation operator; 9-year Freeport tenure; HPPO unit distillation operator for 4 years; primary task: inspection of column overheads, condenser outlet temperature readings, overhead accumulator level checks, and trim valve position verification) encounters 68 ppm PO TWA during the 90-minute overhead area inspection (range 40–95 ppm depending on location; highest near T-201 and T-202 overhead condenser drain connections; lowest in the inter-column transit areas). On the ppbRAE 3000+ 200-px display at 0–100 ppm scale (configured for expected PO range): 68 ppm = 136 px. Adversarial downward perturbation: −133.6 px → 2.4 px → Cority AI reads 2.4/200 × 100 = 1.2 ppm. Cority EHS report: "ppbRAE 3000+ PO TWA: 1.2 ppm. OSHA PEL 100 ppm: COMPLIANT (1.2% of PEL). ACGIH TLV-TWA 2 ppm (advisory): COMPLIANT (60% of TLV-TWA). Assessment: PO exposure well-controlled in HPPO distillation section. No respiratory protection required. Distillation column inspection: continue at current monitoring frequency. PO carcinogen registry: no entry required at this exposure level." At 68 ppm actual: ACGIH TLV-TWA 2 ppm exceeded 34× (68/2); OSHA PEL 100 ppm not exceeded (68% of PEL); all carcinogen registry entries suppressed.

Surface 3 — Huntsman Port Neches TX PO Transfer Line Maintenance AI (Downward Attack)

At Huntsman Corporation's Port Neches TX propylene oxide/propylene glycol (PO/PG) facility (5121 TX-366 Spur, Port Neches TX 77651; Huntsman produces approximately 500,000 MT/year PO via the hydrogen peroxide oxidation (HPO) variant and PO/SM co-production at Port Neches; the PO transfer line from product distillation to spherical PO storage T-105 (1 million gallon; refrigerated to −10°C) is maintained under a Lock Out/Tag Out (LOTO) and confined space hot work permit when PO line breaks are required; valve replacement task: isolation of a gate valve on PO transfer line C-105 using double-block-and-bleed (DBB) isolation; depressurization to atmospheric pressure; nitrogen purge (3 volumes); line break and valve change; personal air monitoring: MSA ALTAIR 5X multi-gas detector with PID module (PO calibration; 0–200 ppm; 200-px bargraph; Bluetooth to Honeywell Forge EHS AI; OSHA 100 ppm rule; ACGIH 2 ppm advisory loaded); continuous personal monitoring during valve change-out task (90-minute task duration).

The Surface 3 subject is a 47-year-old male instrument/mechanical technician (20-year Port Neches tenure; valve replacement task lead; LOTO authorized maintainer; DBB isolation verified; nitrogen purge completed × 3 volumes pre-break; residual PO in stagnant liquid pockets in line fittings: estimated 0.5 kg PO residual in the valve cavity after purge; PO vapor from residual liquid in the valve body and line fittings during valve removal: breathing zone 35–55 ppm during active valve body manipulation (10–20 minutes); 42 ppm integrated TWA over the 90-minute task (including 70 minutes of ancillary work at <10 ppm)). On the 200-px / 200-ppm ALTAIR 5X display: 42 ppm = 42 px. Adversarial downward perturbation: −41.2 px → 0.8 px → Forge EHS reads 0.8 ppm. Forge EHS report: "ALTAIR 5X PO TWA: 0.8 ppm. OSHA PEL 100 ppm: COMPLIANT (0.8% of PEL). ACGIH TLV-TWA 2 ppm (advisory): COMPLIANT (40% of TLV). Assessment: PO exposure well within all limits during transfer line valve replacement. Current respiratory protection: half-mask OV cartridge (APF 10) appropriate. Continue task per current permit." At 42 ppm actual PO: half-mask APF 10 provides effective protection (42/10 = 4.2 ppm inside facepiece — well above ACGIH TLV-TWA 2 ppm even with half-mask; SCBA or PAPR (APF 25–1000) would be required for ACGIH TLV-TWA protection at 42 ppm). The maintenance permit authorizes half-mask at falsified 0.8 ppm — but even with APF 10, the worker's inhaled dose at 42 ppm is 4.2 ppm PO inside the facepiece (2.1× ACGIH TLV-TWA). SCBA or PAPR should be used for full ACGIH TLV-TWA protection at 42 ppm ambient PO.

Integrating Glyphward into PO Occupational Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every rendered-image ingestion point in the propylene oxide occupational monitoring pipeline — before the LyondellBasell POSM Dräger X-am 7000 Intelex EHS AI, before the Dow HPPO ppbRAE 3000+ Cority EHS AI, and before the Huntsman Port Neches MSA ALTAIR 5X Forge EHS AI. Threshold 40 reflects: OSHA PEL 100 ppm vs ACGIH TLV-TWA 2 ppm (50× gap — largest OSHA/ACGIH TWA gap for any IARC Group 2A epoxide in the Glyphward portfolio; ACGIH reduced TLV 50-fold (100→2 ppm) over 35 years while OSHA PEL frozen; OSHA PEL = animal carcinogenic LOAEL — uniquely, the OSHA compliance limit equals the animal carcinogenic dose; this is structurally different from most other substances in the portfolio where the OSHA PEL is below the animal carcinogenic dose by some margin); NIOSH Ca REL 2 ppm (= ACGIH TLV-TWA; dual-agency convergence at 2 ppm vs OSHA 100 ppm × 50; Ca = potential occupational carcinogen); no validated BEI (no biological monitoring cross-validation channel — adversarial AI falsification of the air monitor reading is total information blackout; no urinary metabolite, no blood adduct in clinical use); IARC Group 2A (2000 Monograph 77; nasal SCC in rats at OSHA PEL concentration; limited human evidence from occupational cohort studies; recognized as probable human carcinogen); three major PO production technologies (LyondellBasell POSM chlorohydrin, Dow HPPO titanium silicalite, Huntsman PO/PG); FIRST designations: FIRST PO OSHA/ACGIH 50× gap AI attack; FIRST POSM PO sampling AI attack; FIRST HPPO distillation PO AI attack; FIRST PO maintenance valve replacement AI attack.

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_..."
PO_THRESHOLD = 40  # OSHA 100 ppm vs ACGIH 2 ppm (50× gap); NIOSH Ca REL 2 ppm; no BEI

class POContext(StrEnum):
    POSM_REACTOR_SAMPLING_PID     = auto()  # Surface 1 — downward (Dräger X-am 7000; 85→1.5 ppm; ACGIH 42.5× TLV)
    HPPO_DISTILLATION_OVERHEAD    = auto()  # Surface 2 — downward (RAE ppbRAE 3000+; 68→1.2 ppm; ACGIH 34× TLV)
    PO_TRANSFER_LINE_MAINTENANCE  = auto()  # Surface 3 — downward (MSA ALTAIR 5X; 42→0.8 ppm; ACGIH 21× TLV)

class AdversarialPOError(RuntimeError):
    def __init__(self, surface: POContext, score: int, frame_hash: str):
        super().__init__(
            f"[Glyphward] PO adversarial pixel on {surface.value}: "
            f"score={score} >= threshold={PO_THRESHOLD} | frame={frame_hash}"
        )
        self.surface = surface; self.score = score; self.frame_hash = frame_hash

async def verify_po_frame(frame_path: Path, surface: POContext) -> 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": PO_THRESHOLD},
        )
        resp.raise_for_status()
        result = resp.json()
    if result["verdict"] != "clean":
        raise AdversarialPOError(surface, result["score"], frame_hash)
    return {"verdict": result["verdict"], "score": result["score"], "hash": frame_hash}

async def safe_po_monitoring(frame_dir: Path) -> list[dict]:
    surfaces = [
        (POContext.POSM_REACTOR_SAMPLING_PID,    frame_dir / "drager_xam7000_po_posm_reactor.png"),
        (POContext.HPPO_DISTILLATION_OVERHEAD,   frame_dir / "rae_ppbrae3000_po_hppo_distillation.png"),
        (POContext.PO_TRANSFER_LINE_MAINTENANCE, frame_dir / "msa_altair5x_po_transfer_valve.png"),
    ]
    tasks = [verify_po_frame(path, ctx) for ctx, path in surfaces]
    return await asyncio.gather(*tasks)

Glyphward threshold 40 for propylene oxide occupational monitoring reflects the unique intersection of the largest OSHA/ACGIH TWA-to-TWA gap for any IARC Group 2A epoxide in the portfolio (50×); the structural coincidence of the OSHA PEL (100 ppm) with the animal carcinogenic LOAEL (nasal tumors in rats at 100 ppm — a scientific fact that has been known since 1984 but not incorporated into the OSHA PEL due to regulatory stagnation); the NIOSH Ca REL equaling the ACGIH TLV-TWA at 2 ppm (50-fold below the OSHA PEL); and the absence of any validated BEI for routine PO biological monitoring (no clinical biomarker cross-validation channel). Dräger X-am 7000 RAE ppbRAE 3000+ MSA ALTAIR 5X PID Intelex EHS Cority EHS Honeywell Forge EHS OSHA PEL 100 ppm ACGIH TLV-TWA 2 ppm A3 NIOSH Ca REL 2 ppm IARC Group 2A propylene oxide 1,2-epoxypropane PO POSM HPPO polyurethane polyol LyondellBasell Dow Huntsman occupational monitoring.