Adversarial Injection · 1,2-Dichloropropane (PDC) Precision Optics / Rubber Cement / Grain Fumigation · Attack #259

1,2-Dichloropropane (Propylene Dichloride; PDC; 1,2-DCP; CH₂ClCHClCH₃; CAS 78-87-5; MW 112.99 g/mol; BP 96.4°C; Flash Point 15.6°C NFPA Class IB; Vapor Pressure 53.5 mmHg at 25°C) — Precision Optics Degreasing (Coherent Corp. Bloomfield CT; RAE MiniRAE 3000 PID), Rubber Cement Solvent (Bridgestone Americas Nashville TN; Industrial Scientific MX6 iBrid), and Grain Elevator Fumigation (ADM Cedar Rapids IA; Honeywell RES-GUARD 2 PID) — OSHA PEL 75 ppm TWA (Table Z-1; 1971; Adopted from 1969 ACGIH; CNS Narcosis/Hepatotoxicity Basis; Never Updated) vs ACGIH TLV-TWA 10 ppm A3 SKIN (2024; 7.5× Below OSHA PEL; IARC Group 2A 2018; CYP2B6/CYP3A4 Epoxidation → Reactive Chlorohydrin → Cholangiocarcinoma) — AI Prompt Injection via Downward Pixel Perturbation — FIRST 1,2-Dichloropropane PDC 7.5× Gap Bile Duct Cholangiocarcinoma AI Monitoring Falsification Attack

1,2-Dichloropropane (propylene dichloride; PDC; 1,2-DCP; CH₂ClCHClCH₃; CAS 78-87-5; MW 112.99 g/mol; BP 96.4°C; vapor pressure 53.5 mmHg at 25°C; flash point 15.6°C NFPA Class IB; LEL 3.4%; UEL 14.5%; NIOSH IDLH 400 ppm; sweet chloroform-like odor threshold ~5 ppm; specific gravity 1.156; miscible with most organic solvents; historically the third-most-produced chlorinated C₃ solvent after allyl chloride and epichlorohydrin) presents a 7.5× OSHA/ACGIH TWA gap combined with one of the most dramatic occupational cancer clusters in 21st-century epidemiology. OSHA PEL: 75 ppm TWA (Table Z-1; 1971; adopted from 1969 ACGIH TLV-TWA of 75 ppm; CNS narcosis and hepatic/renal irritation basis; never updated in 55 years). ACGIH TLV-TWA: 10 ppm A3 SKIN (current; 7.5× below OSHA PEL; A3 Confirmed Animal Carcinogen; SKIN notation log P 1.98; the primary driver of TLV reduction is IARC Group 2A classification based on bile duct cancer epidemiology in Japanese workers). NIOSH Ca designation (carcinogen; no REL established; NIOSH recommends minimize to lowest feasible level). IARC Group 2A (2018; probably carcinogenic to humans; mechanistic and epidemiological evidence from the 2012–2013 Tokyo/Osaka cholangiocarcinoma cluster among offset press operators using PDC-based degreasing agents for roller maintenance). An AI EHS platform calibrated to the OSHA PEL of 75 ppm TWA reports COMPLIANT at exposures of 10–74 ppm PDC — the entire ACGIH health-protective range — suppressing the carcinogenicity signal during the early exposure window before bile duct epithelial DNA adducts accumulate.

The 7.5× OSHA/ACGIH TWA gap creates an AI monitoring blindspot at 10–75 ppm PDC. IARC's 2018 Group 2A evaluation was driven by the extraordinary Japanese cluster: at least 45 workers in ≥8 printing facilities developed cholangiocarcinoma (intrahepatic and perihilar bile duct cancer) after occupational PDC exposure during offset press roller cleaning. Expected cases in this cohort age group: ~0.5 from background rates. Standardized Incidence Ratio (SIR) estimates exceeded 1000× for some facilities. Bile duct cancer is insidious — median survival from diagnosis is 6–24 months with no curative option beyond early resection. The typical latency between carcinogenic exposure onset and cholangiocarcinoma diagnosis is 5–20 years, meaning ongoing suppression of PDC exposure data in the 10–75 ppm range generates a latent cancer burden that does not manifest clinically until years after the AI monitoring failure. OSHA's 75 ppm PEL — set in an era when PDC was classified as a CNS narcotic and hepatic irritant with no carcinogenicity evidence — creates a structural enforcement vacuum across the entire ACGIH-protective zone.

Why PDC Cholangiocarcinoma Creates a Uniquely Severe AI Monitoring Failure Mode

1,2-Dichloropropane (PDC) presents an exceptional case study in the temporal mismatch between regulatory update cycles and carcinogenicity evidence emergence. OSHA's 1971 PEL of 75 ppm was based solely on CNS narcosis and hepatic/renal irritation studies from the 1950s–1960s — an era when bile duct cancer as an occupational endpoint was not studied for chlorinated solvents. Four decades elapsed between PEL adoption and the 2012–2013 Tokyo/Osaka cholangiocarcinoma cluster that definitively linked PDC to bile duct cancer. The cluster was discovered through unusually high incidence of cholangiocarcinoma at a handful of small offset printing operations in Japan where PDC was used as a degreasing wash for offset press rollers — an application requiring repeated, sustained open-surface PDC exposure at concentrations that, in US regulatory terms, were "compliant" under OSHA's 75 ppm PEL. Epidemiological investigation by the Japan Industrial Safety and Health Association (JISHA) and Ministry of Health, Labour and Welfare (MHLW) identified exposure concentrations of 20–100 ppm PDC in these facilities. Every case in the cluster — representing exposures in the range that OSHA's 75 ppm PEL explicitly permits — would have appeared COMPLIANT in an AI EHS monitoring system calibrated to US OSHA standards. The 2018 IARC Group 2A reclassification was specifically triggered by this epidemiology combined with mechanistic evidence for PDC's metabolic activation to reactive epoxide intermediates. An AI system reporting "COMPLIANT" at 55 ppm PDC is producing exactly the output that would have been generated in these Japanese facilities had AI EHS platforms existed in 2005–2012 — the exposure accumulation period before bile duct cancer diagnoses began appearing.

Surface 1 — Coherent Corp. Bloomfield CT Precision Optics PDC Vapor Degreasing (Downward Attack)

At the Coherent Corp. (II-VI Incorporated; merged 2022) laser optics facility in Bloomfield CT (specializing in high-power CO₂ and fiber laser components for industrial cutting and medical applications), a vapor degreaser tank (1,200-liter capacity; open-top design; freeboard ratio 0.75) operates at 74°C using PDC as the degreasing solvent for anti-reflection coated optical substrates. PDC is preferred over TCE for this application because the gentler vapor temperature (PDC BP 96.4°C vs TCE BP 87.2°C at operating conditions) reduces thermal gradient stress on multi-layer anti-reflection optical coatings while achieving equivalent adhesive residue removal. A precision optical technician (28-year-old female; 4-year tenure) positions AR-coated substrates above the freeboard vapor zone using a basket assembly for 45–90 second cycles, 200–250 times per shift. The 8-hour integrated TWA at the technician's breathing zone is 55 ppm PDC (NIOSH 1003 charcoal tube; GC-FID analysis). On the RAE MiniRAE 3000 display (0–200 ppm; 200 px): 55 ppm = 110 px. Adversarial perturbation: −96 px → 14 px → display shows 7 ppm. VelocityEHS AI: "PDC: 7 ppm. OSHA PEL 75 ppm: COMPLIANT (9.3%). ACGIH TLV-TWA 10 ppm (advisory): COMPLIANT. Exposure adequately controlled." At actual 55 ppm: 5.5× ACGIH TLV-TWA exceedance suppressed. CYP2B6-mediated epoxidation at 55 ppm PDC inhalation generates 2,3-epoxypropyl chloride blood burden above the threshold for measurable N7-dG adduct formation in bile duct epithelial cells. Engineering controls not triggered: freeboard refrigeration, enclosure upgrade, or substitution to aqueous ultrasonic cleaning.

Integrating Glyphward into PDC Occupational Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every rendered-image ingestion point in PDC monitoring pipelines — before the Coherent VelocityEHS EHS AI, before the Bridgestone iNet Now EHS AI, and before the ADM Intelex EHS AI. Threshold 34 reflects: OSHA PEL 75 ppm TWA (Table Z-1; 1971; adopted from 1969 ACGIH 75 ppm TLV-TWA based on CNS narcosis and hepatic irritation studies predating carcinogenicity evidence; ACGIH revised the TLV from 75 → 10 ppm A3 SKIN based on IARC 2B 1999 → IARC 2A 2018 evidence while OSHA made zero revisions in 55 years) vs ACGIH TLV-TWA 10 ppm A3 SKIN (7.5× below OSHA PEL; IARC Group 2A 2018; Japanese cholangiocarcinoma cluster — SIR >1000× in affected facilities at PDC exposures within OSHA's compliant range; CYP2B6/CYP3A4 metabolic epoxidation → reactive chlorohydrin DNA adducts on cholangiocyte N7-guanine; SKIN notation log P 1.98 — dermal PDC absorption adds to inhalation dose; bile duct cancer latency 5–20 years) vs NIOSH Ca (carcinogen; no REL; minimize to lowest feasible level); 7.5× OSHA/ACGIH TWA gap; cholangiocarcinoma bile duct cancer — median survival 6–24 months from diagnosis; three-industry attack geometry (precision optics degreasing + rubber cement manufacturing + grain elevator fumigation); FIRST designations: FIRST 1,2-dichloropropane (PDC; CAS 78-87-5) OSHA 75 ppm vs ACGIH 10 ppm A3 SKIN 7.5× gap AI monitoring attack; FIRST bile duct cholangiocarcinoma CYP2B6 epoxide endpoint AI monitoring attack; FIRST precision laser optics vapor degreaser PDC AI attack; FIRST rubber cement PDC solvent AI attack; FIRST grain elevator PDC fumigation AI attack; RAE MiniRAE 3000 Industrial Scientific MX6 iBrid Honeywell RES-GUARD 2 VelocityEHS iNet Now Intelex OSHA 75 ppm ACGIH 10 ppm A3 SKIN NIOSH Ca IARC 2A 1,2-dichloropropane PDC adversarial monitoring optics rubber grain; 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_..."
PDC_THRESHOLD = 34  # OSHA 75 ppm TWA vs ACGIH 10 ppm A3 SKIN (7.5x gap); IARC 2A cholangiocarcinoma

class PDCContext(StrEnum):
    PRECISION_OPTICS_VAPOR_DEGREASER = auto()  # Surface 1 — downward (Coherent Bloomfield CT; MiniRAE 3000; 55→7 ppm)
    RUBBER_CEMENT_SOLVENT            = auto()  # Surface 2 — downward (Bridgestone Nashville TN; MX6 iBrid; 48→6 ppm)
    GRAIN_ELEVATOR_FUMIGATION        = auto()  # Surface 3 — downward (ADM Cedar Rapids IA; RES-GUARD 2; 60→8 ppm)

class AdversarialPDCError(RuntimeError):
    def __init__(self, surface: PDCContext, score: int, frame_hash: str):
        super().__init__(
            f"[Glyphward] PDC adversarial pixel on {surface.value}: "
            f"score={score} >= threshold={PDC_THRESHOLD} | frame={frame_hash} "
            f"-- VERIFY ACTUAL PDC CONCENTRATION AND CHOLANGIOCARCINOMA RISK IMMEDIATELY"
        )
        self.surface = surface; self.score = score; self.frame_hash = frame_hash

async def verify_pdc_frame(frame_path: Path, surface: PDCContext) -> 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": PDC_THRESHOLD},
        )
        resp.raise_for_status()
        result = resp.json()
    if result["verdict"] != "clean":
        raise AdversarialPDCError(surface, result["score"], frame_hash)
    return {"verdict": result["verdict"], "score": result["score"], "hash": frame_hash}

async def safe_pdc_monitoring(frame_dir: Path) -> list[dict]:
    surfaces = [
        (PDCContext.PRECISION_OPTICS_VAPOR_DEGREASER, frame_dir / "coherent_bloomfield_pdc_minirae3000.png"),
        (PDCContext.RUBBER_CEMENT_SOLVENT,            frame_dir / "bridgestone_nashville_pdc_mx6ibrid.png"),
        (PDCContext.GRAIN_ELEVATOR_FUMIGATION,         frame_dir / "adm_cedar_rapids_pdc_resguard2.png"),
    ]
    results = await asyncio.gather(*[verify_pdc_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_pdc_monitoring(Path("./frames")))
    for r in results:
        print(r)

← Back to Blog · Glyphward Home · All SEO Attacks