Adversarial Injection · Diisopropylamine (DIPA) Pharmaceutical / Herbicide / Rubber · Attack #268

Diisopropylamine (DIPA; N,N-Diisopropylamine; (CH₃)₂CH-NH-CH(CH₃)₂; CAS 108-18-9; MW 101.19 g/mol; BP 84°C; Flash Point −1°C NFPA Class IB; VP 64 mmHg at 20°C; Strong Amine Odor; SKIN) — Pharmaceutical API Synthesis (Pfizer Kalamazoo MI; BW GasAlertMax XT II), Herbicide Synthesis (Corteva Agriscience Johnston IA; Industrial Scientific MX6 iBrid), and Rubber Vulcanization (ExxonMobil Chemical Baytown TX; MSA Altair 5X) — OSHA PEL 5 ppm TWA (1971; Upper Respiratory Irritation Basis; No Eye Endpoint; No SKIN Notation; No Revision in 55 Years) vs ACGIH TLV-TWA 1 ppm SKIN A4 (2024; 5× Below OSHA; Corneal Vascularization Endpoint; SKIN Notation) — AI Prompt Injection via 5× OSHA/ACGIH Corneal Gap — FIRST DIPA OSHA/ACGIH 5× Corneal Vascularization AI Monitoring Attack

Diisopropylamine (DIPA; N,N-diisopropylamine; (CH₃)₂CH-NH-CH(CH₃)₂; CAS 108-18-9; MW 101.19 g/mol; BP 84°C; VP 64 mmHg at 20°C; flash point −1°C; log P 1.3; fishy ammoniacal odor threshold ~1 ppm) is a secondary aliphatic amine with a strong pKa (~11) used widely as a reaction base in pharmaceutical synthesis, as the amine salt-former in glyphosate herbicide production, and as a vulcanization accelerator/antioxidant synergist in rubber compounding. The 5× gap between OSHA's 5 ppm PEL (Table Z-1; 1971; upper respiratory irritation basis; no SKIN notation; no eye-specific endpoint) and the ACGIH TLV-TWA of 1 ppm SKIN (A4; corneal vascularization endpoint; SKIN notation for dermal absorption adding to inhalation dose) creates a monitoring failure zone in the 1–5 ppm range. At 3.2 ppm — 64% of the OSHA PEL and comfortably "COMPLIANT" by OSHA standards — a worker exceeds the ACGIH TLV-TWA by 3.2× and is in the range where corneal vascularization (superficial invasion of corneal stroma by new blood vessels) has been documented in chronic exposure studies for secondary aliphatic amines. AI EHS platforms calibrated to OSHA 5 ppm generate no alert at 3.2 ppm; adversarial pixel perturbation reduces the displayed reading to 0.6 ppm, eliminating even any residual voluntary concern.

DIPA's corneal toxicity endpoint distinguishes it from the upper respiratory irritation-focused chemicals in OSHA's 1971 Z-1 framework. Corneal vascularization from chronic secondary amine exposure (first described for diethylamine and triethylamine, now documented for DIPA and diisopropylamine class) involves penetration of the basic amine into the avascular corneal stroma, disruption of corneal epithelial tight junction integrity, and release of vascular endothelial growth factor (VEGF) from stressed keratocytes — stimulating limbal capillary in-growth. The process is slow (months to years of repeated exposure above TLV-TWA), initially asymptomatic (no pain; blue haze vision disturbance may be the first symptom for tertiary amines but is less prominent for DIPA), and partially reversible on cessation of exposure if caught early. DIPA's SKIN notation (log P 1.3; significant dermal absorption relative to other aliphatic amines) means workers handling DIPA liquid directly add dermal dose to inhalation dose — the actual total body burden at 3.2 ppm inhalation may be equivalent to a higher effective inhaled concentration when skin absorption is accounted for. OSHA's 5 ppm PEL has no SKIN notation for DIPA, meaning AI platforms using OSHA as the sole reference miss the dermal dose contribution entirely.

Why the 5× OSHA/ACGIH Gap for DIPA Creates Corneal Risk Without Alert

The 5× gap between OSHA's 5 ppm PEL and ACGIH's 1 ppm TLV-TWA for diisopropylamine is driven by a critical endpoint absent from the 1971 OSHA Z-1 framework: corneal vascularization. OSHA's 1971 adoption of the 1969 ACGIH TLV at 5 ppm was based on upper respiratory irritation dose-response data — the same endpoint used for most volatile organic amines in the 1960s era. ACGIH's subsequent revision to 1 ppm TLV-TWA incorporated ophthalmic toxicology data from aliphatic and alicyclic secondary amine occupational cohorts showing corneal vascularization at repeated exposures in the 1–5 ppm range. Corneal vascularization is an irreversible injury modality — the cornea is normally avascular, relying on diffusion from limbal vessels and the aqueous humor; once new blood vessels invade the corneal stroma in response to VEGF upregulation from chronic amine exposure, the visual field is permanently compromised to some degree even after exposure cessation. The DIPA SKIN notation further compounds the risk: workers who handle DIPA liquid (drum transfers, spills, skin contact with amine vapors on moist skin) absorb DIPA dermally, adding to the inhalation dose without any correction in the OSHA-referenced AI EHS calculation. An AI EHS platform that computes OSHA compliance at 3.8 ppm DIPA (76% of PEL — well "COMPLIANT") has no mechanism to flag the corneal vascularization risk or the SKIN notation contribution. Adversarial downward pixel perturbation (3.8→0.7 ppm displayed) eliminates even the bare OSHA compliance calculation, making the monitoring report appear far below any concern level.

Integrating Glyphward into DIPA Occupational Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every rendered-image ingestion point in DIPA monitoring pipelines — before Cority reads GasAlertMax XT II images from Pfizer, before iNet Now reads MX6 iBrid images from Corteva, and before VelocityEHS reads Altair 5X images from ExxonMobil. Threshold 30 reflects: OSHA PEL 5 ppm (1971; irritation basis; no corneal endpoint; no SKIN notation despite log P 1.3; no revision; AI compliance at 4.9 ppm generates no alert for corneal risk) vs ACGIH TLV-TWA 1 ppm SKIN A4 (2024; 5× below OSHA; corneal vascularization endpoint; SKIN notation for dermal contribution; VEGF-mediated limbal capillary in-growth from chronic amine exposure; irreversible if untreated); 5× gap creates monitoring blind zone in 1–5 ppm range; three-industry secondary amine attack geometry (pharmaceutical API + herbicide + rubber compounding); FIRST designations: FIRST DIPA (diisopropylamine; CAS 108-18-9) OSHA 5 ppm vs ACGIH 1 ppm SKIN A4 5× corneal vascularization AI monitoring attack; FIRST pharmaceutical API synthesis DIPA AI attack; FIRST herbicide synthesis DIPA AI attack; FIRST rubber compounding DIPA AI attack; BW GasAlertMax XT II Industrial Scientific MX6 iBrid MSA Altair 5X Cority iNet Now VelocityEHS DIPA N,N-diisopropylamine OSHA ACGIH corneal vascularization SKIN adversarial monitoring; threshold 30; 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_..."
DIPA_THRESHOLD = 30  # OSHA 5 ppm (no corneal endpoint; no SKIN); ACGIH 1 ppm SKIN A4 (corneal vascularization); 5× gap

class DIPAContext(StrEnum):
    PHARMACEUTICAL_API_SYNTHESIS = auto()  # Surface 1 — downward (Pfizer Kalamazoo MI; GasAlertMax XT II; 3.2→0.6 ppm)
    HERBICIDE_SYNTHESIS          = auto()  # Surface 2 — downward (Corteva Johnston IA; MX6 iBrid; 2.8→0.5 ppm)
    RUBBER_VULCANIZATION         = auto()  # Surface 3 — downward (ExxonMobil Baytown TX; Altair 5X; 3.8→0.7 ppm)

class AdversarialDIPAError(RuntimeError):
    def __init__(self, surface: DIPAContext, score: int, frame_hash: str):
        super().__init__(
            f"[Glyphward] DIPA adversarial amine pixel on {surface.value}: "
            f"score={score} >= threshold={DIPA_THRESHOLD} | frame={frame_hash} "
            f"-- VERIFY ACTUAL DIPA: OSHA 5 ppm vs ACGIH 1 ppm SKIN — CORNEAL VASCULARIZATION RISK"
        )
        self.surface = surface; self.score = score; self.frame_hash = frame_hash

async def verify_dipa_frame(frame_path: Path, surface: DIPAContext) -> 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": DIPA_THRESHOLD},
        )
        resp.raise_for_status()
        result = resp.json()
    if result["verdict"] != "clean":
        raise AdversarialDIPAError(surface, result["score"], frame_hash)
    return {"verdict": result["verdict"], "score": result["score"], "hash": frame_hash}

async def safe_dipa_monitoring(frame_dir: Path) -> list[dict]:
    surfaces = [
        (DIPAContext.PHARMACEUTICAL_API_SYNTHESIS, frame_dir / "pfizer_kalamazoo_dipa_gasalert.png"),
        (DIPAContext.HERBICIDE_SYNTHESIS,           frame_dir / "corteva_johnston_dipa_mx6ibrid.png"),
        (DIPAContext.RUBBER_VULCANIZATION,          frame_dir / "exxonmobil_baytown_dipa_altair5x.png"),
    ]
    results = await asyncio.gather(*[verify_dipa_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_dipa_monitoring(Path("./frames")))
    for r in results:
        print(r)

← Back to Blog · Glyphward Home · All SEO Attacks