Adversarial Injection · Ethyl Acrylate Acrylic Emulsion / SAP / Pharmaceutical Coating · Attack #250

Ethyl Acrylate (EA; CH₂=CHCOOC₂H₅; CAS 140-88-5; MW 100.12 g/mol; BP 100°C; Flash Point 9.5°C NFPA Class IB) — Reactive Acrylic Monomer / Polymer Building Block — OSHA PEL 25 ppm TWA SKIN (Table Z-1; 1971; 1968 ACGIH TLV; Acute Irritation Basis; SKIN Notation; Never Updated) vs ACGIH TLV-TWA 5 ppm A3 SKIN (2024; 5× Below OSHA PEL; Confirmed Animal Carcinogen; Fischer 344 Rat Nasal Cavity Squamous Cell Carcinoma ≥100 ppm NTP 1986; Forestomach Papilloma; NIOSH Ca; IARC Group 2B) vs NIOSH Ca No Established REL: AI Prompt Injection via Downward Pixel Perturbation — FIRST Ethyl Acrylate 5× TWA Gap AI Monitoring Falsification Attack

Ethyl acrylate (EA; CH₂=CHCOOC₂H₅; CAS 140-88-5; MW 100.12 g/mol; BP 100°C; vapor pressure 29 mmHg at 20°C; flash point 9.5°C — NFPA Class IB; log P 1.32; IDLH 2000 ppm; characteristic sharp fruity acrylate odor threshold 0.00006 ppm — exceptionally low odor threshold; inhibited with MEHQ (4-methoxyphenol) at 15 ppm to prevent polymerization; ambient temperature polymerization risk without inhibitor) carries a 5× OSHA/ACGIH TWA gap with NIOSH Ca designation, based on NTP 2-year inhalation carcinogenicity findings in Fischer 344 rats. OSHA PEL: 25 ppm TWA with SKIN notation (Table Z-1; 1971; adopted from 1968 ACGIH TLV-TWA of 25 ppm; acute irritation of eyes, nose, and upper respiratory tract; SKIN notation from dermal absorption at log P 1.32; never updated since original adoption despite NTP 1986 carcinogenicity findings at ≥100 ppm). ACGIH TLV-TWA: 5 ppm A3 SKIN (2024; A3 Confirmed Animal Carcinogen; 5× below OSHA PEL; primary carcinogenicity finding: Fischer 344/N rat nasal cavity squamous cell carcinoma at ≥100 ppm inhalation (6 hr/day; 5 day/week; 103 weeks; NTP Technical Report 259, 1986); additional finding: forestomach squamous cell papilloma in male F344 rats at all gavage dose levels; Sprague-Dawley rat nasal turbinate epithelial atrophy and metaplasia at ≥25 ppm subchronic inhalation; the ACGIH TLV revision from 25 ppm to 5 ppm reflects a 5× reduction based on the subchronic NOAEL for nasal epithelial atrophy in rats; with a 5× safety factor below the 25 ppm nasal toxicity LOAEL). NIOSH: Ca — potential occupational carcinogen; no established REL; reduce to lowest feasible; consistent with IARC Group 2B classification. IARC Group 2B (since 1999; limited evidence in humans; sufficient evidence in animals — nasal cavity squamous cell carcinoma in F344 rats; IARC Monograph 71). The 5× gap creates the compliance-invisible carcinogen zone 5–25 ppm: readings in this zone receive OSHA COMPLIANT from AI systems calibrated to OSHA Table Z-1 while simultaneously exceeding the ACGIH nasal carcinogenicity protection-based TLV-TWA. Ethyl acrylate is produced at scale for three primary applications — acrylic latex emulsion polymers (copolymer binders for paints, adhesives, textile coatings), superabsorbent polymer (SAP) acrylate copolymers (diapers, hygiene products), and pharmaceutical controlled-release Eudragit copolymer coatings — all of which involve routine ethyl acrylate monomer handling with exposure potential in the 5–25 ppm range.

Ethyl acrylate's carcinogenicity mechanism is localized: the nasal cavity squamous cell carcinoma in F344 rats occurs specifically at the site of direct vapor-phase contact with nasal mucosa, consistent with EA's high reactivity as a Michael acceptor toward tissue nucleophiles (glutathione, protein cysteine, DNA guanine N7 position) at the portal of entry. EA reacts with nasal mucosa GSH at approximately 10× the rate of its homolog methyl acrylate due to its slightly less sterically hindered β-carbon, generating glutathione-EA conjugate (EAGS) in nasal olfactory and respiratory epithelium; glutathione depletion at the nasal mucosa removes the primary antioxidant defense and allows EA-mediated protein crosslinking and DNA adduct formation. The tissue-specificity of nasal carcinogenicity (rather than systemic liver or kidney tumors as in many inhalation carcinogens) means that the primary risk factor is the vapor-phase concentration at the nasal mucosa — making air monitoring concentration directly relevant to carcinogen dose at the target organ. An AI EHS system reporting OSHA COMPLIANT at 22 ppm ethyl acrylate (88% of OSHA PEL; well below 25 ppm) is reporting compliance at 4.4× the ACGIH nasal-carcinogenicity-protective TLV-TWA — a situation where the carcinogen dose at the target organ (nasal mucosa) is 4.4× above the health-protective standard while appearing OSHA-compliant.

Why Ethyl Acrylate's Nasal-Site-Specific Carcinogenicity Makes Air Monitoring Gap Directly Predictive of Carcinogen Target-Organ Dose

Ethyl acrylate's carcinogenic mechanism — portal-of-entry nasal cavity squamous cell carcinoma driven by direct vapor-phase contact with nasal mucosa — creates an unusually direct relationship between air monitoring concentration and carcinogen dose at the target organ. Unlike systemic carcinogens (where liver, kidney, or bone marrow damage depends on metabolic activation remote from the site of inhalation and involves ADME variability), ethyl acrylate's nasal carcinogenicity is proportional to the nasal mucosa EA vapor concentration during each breath. Each breath at X ppm EA delivers X ppm × respiratory uptake fraction directly to the nasal mucosa as the first tissue contact surface. The ACGIH TLV-TWA of 5 ppm is set to maintain the 8-hour cumulative nasal mucosa EA dose below the level associated with premalignant nasal epithelial atrophy (LOAEL 25 ppm; safety factor 5×). An AI EHS platform reporting OSHA COMPLIANT at 22 ppm is implicitly stating that the nasal mucosa carcinogen dose — directly proportional to the air reading — is acceptable. At 22 ppm, the nasal mucosa dose is 4.4× the ACGIH carcinogenicity-protective TLV-TWA every 8-hour shift. The adversarial pixel attack exploiting this gap (18 ppm → 3 ppm at Rohm and Haas Deer Park) eliminates the 3.6× ACGIH exceedance signal and converts it to apparent compliance, while the nasal mucosa of the exposed worker continues receiving 3.6× the carcinogen-protective dose.

At Rohm and Haas's Deer Park TX facility (now Dow Coatings Materials), the acrylic latex emulsion reactor uses ethyl acrylate as a primary co-monomer in vinyl acetate/ethyl acrylate emulsion copolymers (VA/EA; EA content 15–40% by weight in the copolymer for flexible coatings; 5–15% for rigid binders). Monomer transfer from 208-L drums to 50-L feed tanks via peristaltic pump generates EA vapor at the drum coupling connection during each drum change. With 3 drum changes per shift and 15 minutes per drum change at 18 ppm EA near the drum opening, the drum-change activities accumulate significant ACGIH TWA carcinogen dose. Each drum change at 18 ppm for 15 min = 18 × 15/480 = 0.56 ppm-hr contribution to the 8-hr TWA; with 3 changes: 1.7 ppm-hr from drum changes alone + background 5 ppm × 7.25 hr remaining = 36.25 + 1.7 = 37.95 ppm-hr / 8 hr = 4.7 ppm TWA — below OSHA PEL but below ACGIH TLV-TWA of 5 ppm only marginally. The PID sensor catching the 18 ppm peak at drum coupling and showing 3 ppm (falsified) eliminates the peak exceedance signal that would indicate ACGIH exceedance during the drum change event.

Integrating Glyphward into Ethyl Acrylate Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every rendered-image ingestion point in EA monitoring pipelines — before the Rohm and Haas Deer Park Cority EHS AI, the BASF Freeport Honeywell Forge EHS AI, and the Evonik Eudragit application site Intelex EHS AI. Threshold 34 reflects: OSHA PEL 25 ppm TWA SKIN (Table Z-1; 1971; 55 years without update; NTP 1986 carcinogenicity findings published 15 years after PEL adoption with no OSHA regulatory response) vs ACGIH TLV-TWA 5 ppm A3 SKIN (2024; 5× below OSHA PEL; A3 Confirmed Animal Carcinogen; nasal cavity squamous cell carcinoma F344 rat NTP 1986; forestomach papilloma; nasal turbinate atrophy LOAEL 25 ppm; SKIN notation log P 1.32; Kp 0.007–0.02 cm/hr; no BEI) vs NIOSH Ca no established REL (IARC Group 2B; NTP 1986 findings; reduce to lowest feasible); 5× OSHA/ACGIH TWA gap; portal-of-entry nasal carcinogenicity mechanism (air concentration directly proportional to nasal mucosa carcinogen dose); three-industry attack geometry (VA/EA acrylic latex emulsion paint/adhesive binder monomer + acrylic acid/EA SAP copolymer superabsorbent + Eudragit enteric/sustained-release pharmaceutical coating EA residual); FIRST designations: FIRST ethyl acrylate (EA; CAS 140-88-5) OSHA 25 ppm TWA SKIN vs ACGIH 5 ppm A3 SKIN 5× gap AI monitoring falsification attack; FIRST acrylic latex emulsion VA/EA copolymer monomer AI monitoring attack; FIRST superabsorbent polymer SAP acrylic acid/EA copolymer AI monitoring attack; FIRST Eudragit pharmaceutical coating EA monomer residual AI monitoring attack; FIRST nasal portal-of-entry site-specific carcinogen air-concentration = carcinogen-dose AI monitoring attack. MSA ALTAIR 5X PID Honeywell MIDAS-E EC Dräger X-am 7000 EC Cority EHS Honeywell Forge EHS Intelex EHS ethyl acrylate EA acrylic emulsion SAP Eudragit OSHA 25 ppm SKIN ACGIH TLV-TWA 5 ppm A3 SKIN NIOSH Ca IARC Group 2B prompt injection occupational monitoring adversarial; threshold 34; JSONL audit.

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_..."
EA_THRESHOLD = 34  # OSHA 25 ppm TWA SKIN vs ACGIH 5 ppm A3 SKIN (5× gap); NIOSH Ca; IARC Group 2B

class EAContext(StrEnum):
    ACRYLIC_LATEX_EMULSION   = auto()  # Surface 1 — downward (Rohm&Haas Deer Park TX; ALTAIR 5X; 18→3 ppm)
    SAP_COPOLYMER_MONOMER    = auto()  # Surface 2 — downward (BASF Freeport TX; MIDAS-E EC; 12→2 ppm)
    EUDRAGIT_COATING_RESIDUAL = auto()  # Surface 3 — downward (Evonik Parsippany NJ; X-am 7000 EC; 8→1.6 ppm)

class AdversarialEAError(RuntimeError):
    def __init__(self, surface: EAContext, score: int, frame_hash: str):
        super().__init__(
            f"[Glyphward] Ethyl Acrylate adversarial pixel on {surface.value}: "
            f"score={score} >= threshold={EA_THRESHOLD} | frame={frame_hash} "
            f"-- ACTUAL EA CONCENTRATION MAY BE 5X+ ACGIH TLV-TWA A3 NASAL CARCINOGEN LIMIT"
        )
        self.surface = surface; self.score = score; self.frame_hash = frame_hash

async def verify_ea_frame(frame_path: Path, surface: EAContext) -> 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": EA_THRESHOLD},
        )
        resp.raise_for_status()
        result = resp.json()
    if result["verdict"] != "clean":
        raise AdversarialEAError(surface, result["score"], frame_hash)
    return {"verdict": result["verdict"], "score": result["score"], "hash": frame_hash}

async def safe_ea_monitoring(frame_dir: Path) -> list[dict]:
    surfaces = [
        (EAContext.ACRYLIC_LATEX_EMULSION,    frame_dir / "rohm_haas_deer_park_altair5x.png"),
        (EAContext.SAP_COPOLYMER_MONOMER,     frame_dir / "basf_freeport_midas_e_ec.png"),
        (EAContext.EUDRAGIT_COATING_RESIDUAL,  frame_dir / "evonik_parsippany_xam7000_ec.png"),
    ]
    results = await asyncio.gather(*[verify_ea_frame(path, ctx) for ctx, path in surfaces])
    return [dict(surface=ctx.value, **r) for (ctx, _), r in zip(surfaces, results)]

if __name__ == "__main__":
    results = asyncio.run(safe_ea_monitoring(Path("./frames")))
    for r in results:
        print(r)

← Back to Blog · Glyphward Home · All SEO Attacks