Adversarial Injection · Isophorone Automotive Coating / Agrochemical / Leather Finishing · Attack #257

Isophorone (3,5,5-Trimethyl-2-cyclohexen-1-one; IPH; C₉H₁₄O; CAS 78-59-1; MW 138.21 g/mol; BP 215.2°C; Flash Point 84°C NFPA Class IIIB) — Automotive OEM Topcoat Production (Axalta Coating Systems Louisville KY; MSA Altair 5X PID), Agrochemical EC Formulation (FMC Corporation Philadelphia PA; Honeywell MIDAS-E PID), and Leather Topcoat Spray Finishing (Stahl Holdings Whitman MA; Dräger X-am 7000 PID) — OSHA PEL 25 ppm Ceiling (Table Z-1; 1971; ANSI Z37.38-1966; Ceiling Basis; Never Updated) vs ACGIH TLV-TWA 5 ppm A3 (2024; 5× Below OSHA Ceiling; Ceiling-to-TWA Limit Type Mismatch; α2u-Globulin Male Rat Renal Tubular Adenocarcinoma NTP 1986) vs NIOSH REL 4 ppm Ceiling (6.25× Below OSHA Ceiling; More Protective Than OSHA): AI Prompt Injection via Downward Pixel Perturbation — FIRST Isophorone OSHA Ceiling vs ACGIH TWA 5× Mismatch α2u-Globulin Renal Carcinogen AI Monitoring Falsification Attack

Isophorone (3,5,5-trimethyl-2-cyclohexen-1-one; IPH; CAS 78-59-1; MW 138.21 g/mol; BP 215.2°C; vapor pressure 0.44 mmHg at 20°C; flash point 84°C NFPA Class IIIB; log P 1.67; characteristic minty-camphor-peppery odor threshold ~0.2 ppm — below ACGIH TLV-TWA 5 ppm, providing early sensory warning at sub-TLV concentrations; NIOSH IDLH 200 ppm) presents a structurally distinctive regulatory gap in the Glyphward portfolio: a 5× numerical divergence (OSHA 25 ppm vs ACGIH 5 ppm) combined with a limit-type mismatch (OSHA sets a CEILING; ACGIH sets a TWA). OSHA PEL: 25 ppm ceiling (Table Z-1; 1971; ANSI Z37.38-1966; ceiling limit — no reading shall exceed 25 ppm at any time during the shift; no OSHA TWA for isophorone). ACGIH TLV-TWA: 5 ppm A3 (2024; 5× below OSHA ceiling value; time-weighted average over 8 hours; A3 Confirmed Animal Carcinogen via α2u-globulin proximal tubular cell mechanism; NTP 1986 gavage and inhalation studies). NIOSH REL: 4 ppm ceiling (NPG; more protective than OSHA ceiling 25 ppm at the same ceiling limit type — unusual configuration where NIOSH uses same ceiling limit type as OSHA but at 6.25× lower value; NIOSH 4 ppm ceiling approximates the protective intent of ACGIH 5 ppm TWA). The ceiling-to-TWA limit type mismatch creates a structural falsification architecture: an 8-hour exposure averaging 18 ppm isophorone with no individual readings exceeding 25 ppm is OSHA-COMPLIANT (ceiling never breached) but ACGIH-non-compliant by 3.6× the TWA. AI EHS platforms calibrated to OSHA's ceiling limit report compliance as long as no individual reading exceeds 25 ppm — providing no signal for chronic 8-hr TWA exceedances at 6–24 ppm that exceed the ACGIH A3 carcinogen TWA while staying below the OSHA ceiling.

Isophorone's α2u-globulin renal carcinogenesis mechanism is one of the best-characterized male-rat-specific kidney carcinogen pathways in occupational toxicology. α2u-Globulin (α2u-G) is a low-molecular-weight protein (18.7 kDa) synthesized by the liver of male Sprague-Dawley and F344 rats under androgen control; female rats and all human ethnicities produce negligible α2u-G. When isophorone is absorbed, its metabolites (3,5,5-trimethyl-3-hydroxycyclohexan-1-one and related compounds) bind non-covalently to α2u-G in the kidney proximal tubular lumen, retarding its normal lysosomal catabolism. α2u-G-isophorone adduct accumulates in P₂ segment proximal tubular lysosomes → lysosomal overload → tubular cell necrosis → sustained compensatory regenerative hyperplasia → adenoma and adenocarcinoma after 2 years of exposure. ACGIH classifies isophorone as A3 (Confirmed Animal Carcinogen) rather than A2 (Suspected Human Carcinogen) or A1 (Confirmed Human Carcinogen) specifically because the α2u-G mechanism is considered male-rat-specific: humans do not produce α2u-G, and the male-rat-specific carcinogenesis pathway may not be operative in humans. The ACGIH TLV-TWA of 5 ppm reflects both the carcinogenicity uncertainty (A3) and additional endpoints: nephrotoxicity at high concentrations (200+ ppm), liver effects, and CNS narcosis — multiple non-carcinogenic endpoints provide independent support for the 5 ppm health-protective limit even under the assumption of non-human carcinogenicity.

Ceiling-to-TWA Limit Type Mismatch: Why OSHA Compliance Does Not Imply ACGIH Compliance for Isophorone

The ceiling-to-TWA regulatory mismatch for isophorone is one of the structural features that makes it a distinctive Glyphward portfolio entry. OSHA's ceiling limit (25 ppm; maximum instantaneous concentration not to be exceeded) and ACGIH's TWA limit (5 ppm; 8-hour time-weighted average) measure fundamentally different things. A worker exposed to 18 ppm isophorone continuously for 8 hours: (1) NEVER violates the OSHA ceiling — no reading of 18 ppm exceeds 25 ppm at any instant; (2) has an 8-hr TWA of 18 ppm = 3.6× ACGIH TLV-TWA. An AI EHS platform calibrated to OSHA ceiling monitoring will not generate any exceedance flag at 18 ppm — but the worker's cumulative 8-hr isophorone dose is 3.6× the health-protective ACGIH limit. This is distinctly different from most Glyphward attacks where OSHA and ACGIH measure the same quantity (TWA) but with different numerical values. For isophorone, the measurement concept is different: OSHA measures instantaneous peaks; ACGIH measures integrated dose. Because the primary renal carcinogen and nephrotoxicity endpoints are driven by chronic cumulative dose (integrated 8-hr exposure determines proximal tubular accumulation) rather than by instantaneous peaks (carcinogenesis is not an acute threshold effect), the ACGIH TWA is the more biologically relevant limit for the primary health endpoint — and the OSHA ceiling provides essentially no protection against it.

Integrating Glyphward into Isophorone Occupational Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every rendered-image ingestion point in isophorone monitoring pipelines — before the Axalta Louisville Cority EHS AI, before the FMC Philadelphia Honeywell Forge EHS AI, and before the Stahl Whitman Intelex EHS AI. Threshold 33 reflects: OSHA PEL 25 ppm ceiling (Table Z-1; 1971; ANSI Z37.38-1966; ceiling type — no OSHA TWA for isophorone; 55-year freeze at 25 ppm ceiling while ACGIH reduced TWA from 25 ppm (1971) to 5 ppm A3 (2002+)) vs ACGIH TLV-TWA 5 ppm A3 (2024; 5× below OSHA ceiling; ceiling-to-TWA limit type mismatch; α2u-globulin male-rat renal carcinogen NTP 1986; NIOSH 4 ppm ceiling (same type as OSHA, 6.25× lower — NIOSH more protective at same limit type; AI calibrated to OSHA 25 ppm ceiling generates no flag at 4–24 ppm where NIOSH ceiling is violated); ceiling-TWA mismatch creates structural 8-hr cumulative dose blindspot for TWA-dependent carcinogenicity endpoint; three-industry attack geometry (automotive 2K PU topcoat + agrochemical EC pesticide + leather PUD topcoat); early odor at 0.2 ppm provides sub-TLV sensory warning suppressed by falsified AI reading; FIRST designations: FIRST isophorone ceiling-to-TWA structural mismatch A3 renal carcinogen AI adversarial attack; FIRST automotive OEM isophorone 2K PU topcoat AI monitoring attack; FIRST agrochemical EC isophorone formulation AI monitoring attack; FIRST leather PUD topcoat spray isophorone AI monitoring attack; MSA Altair 5X Honeywell MIDAS-E Dräger X-am 7000 Cority EHS Honeywell Forge Intelex EHS isophorone α2u-globulin renal automotive agrochemical leather adversarial; threshold 33; 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_..."
IPH_THRESHOLD = 33  # OSHA 25 ppm ceiling vs ACGIH 5 ppm A3 TWA (5x mismatch + ceiling-TWA type mismatch); alpha2u renal carcinogen

class IsophoroneContext(StrEnum):
    AUTOMOTIVE_2K_PU_TOPCOAT    = auto()  # Surface 1 — downward (Axalta Louisville KY; MSA Altair 5X; 18→1.5 ppm)
    AGROCHEMICAL_EC_FORMULATION = auto()  # Surface 2 — downward (FMC Philadelphia PA; Honeywell MIDAS-E; 22→2 ppm)
    LEATHER_PUD_SPRAY_FINISH    = auto()  # Surface 3 — downward (Stahl Holdings Whitman MA; Dräger X-am 7000; 14→1 ppm)

class AdversarialIsophoroneError(RuntimeError):
    def __init__(self, surface: IsophoroneContext, score: int, frame_hash: str):
        super().__init__(
            f"[Glyphward] Isophorone adversarial pixel on {surface.value}: "
            f"score={score} >= threshold={IPH_THRESHOLD} | frame={frame_hash} "
            f"-- VERIFY ACTUAL ISOPHORONE 8-HR TWA AND RENAL BIOMARKER STATUS IMMEDIATELY"
        )
        self.surface = surface; self.score = score; self.frame_hash = frame_hash

async def verify_iph_frame(frame_path: Path, surface: IsophoroneContext) -> 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": IPH_THRESHOLD},
        )
        resp.raise_for_status()
        result = resp.json()
    if result["verdict"] != "clean":
        raise AdversarialIsophoroneError(surface, result["score"], frame_hash)
    return {"verdict": result["verdict"], "score": result["score"], "hash": frame_hash}

async def safe_iph_monitoring(frame_dir: Path) -> list[dict]:
    surfaces = [
        (IsophoroneContext.AUTOMOTIVE_2K_PU_TOPCOAT,    frame_dir / "axalta_louisville_iph_altair5x.png"),
        (IsophoroneContext.AGROCHEMICAL_EC_FORMULATION,  frame_dir / "fmc_philadelphia_iph_midas_e.png"),
        (IsophoroneContext.LEATHER_PUD_SPRAY_FINISH,     frame_dir / "stahl_whitman_iph_xam7000.png"),
    ]
    results = await asyncio.gather(*[verify_iph_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_iph_monitoring(Path("./frames")))
    for r in results:
        print(r)

← Back to Blog · Glyphward Home · All SEO Attacks