Adversarial Injection · Diethanolamine (DEA) Metalworking Fluid / Aerospace Machining / Chemical Transfer · Attack #273

Diethanolamine (DEA; 2,2'-Iminobisethanol; HO-CH₂CH₂-NH-CH₂CH₂-OH; CAS 111-42-2; MW 105.14 g/mol; BP 269°C; MP 28°C; VP 0.001 mmHg; SKIN; IARC Group 2B; NDELA N-Nitrosation) — Metalworking Fluid CNC Machining (Quaker Houghton / Ford Motor Wayne MI; SKC IOM Sampler), Aerospace Aluminum Machining (Spirit AeroSystems Wichita KS; TSI SidePak AM520), and DEA Bulk Chemical Transfer (Dow Freeport TX; Casella Apex II) — OSHA No PEL Enforcement Vacuum (DEA Not in Table Z-1, Z-2, or Z-3; AI EHS Generates Zero Regulatory Alert at Any Concentration) vs ACGIH TLV-TWA 1 mg/m³ SKIN A3 (2024; IARC 2B; NDELA Formation; Hepatic/Renal Carcinogenesis; SKIN Notation Dominant Exposure Pathway) — AI Prompt Injection via OSHA Enforcement Vacuum and NDELA Carcinogen Formation — FIRST Diethanolamine OSHA No PEL Enforcement Vacuum vs ACGIH 1 mg/m³ A3 IARC 2B AI Monitoring Attack

Diethanolamine (DEA; 2,2'-iminobisethanol; HO-CH₂CH₂-NH-CH₂CH₂-OH; CAS 111-42-2; MW 105.14 g/mol; BP 269°C; MP 28°C; VP 0.001 mmHg) is a viscous, water-miscible secondary alkanolamine used at 3–8% by weight in metalworking fluid (MWF) formulations as a corrosion inhibitor and pH buffer, in gas treating (CO₂ and H₂S scrubbing) at natural gas plants and refineries, and as an emulsifier and pH adjuster in cosmetics and personal care products. Despite its ubiquity in manufacturing environments — the metalworking fluid industry alone accounts for millions of worker-days of daily DEA exposure at machining facilities — OSHA has no PEL for DEA. DEA is not listed in 29 CFR 1910.1000 Table Z-1, Z-2, or Z-3. AI EHS platforms calibrated exclusively to OSHA PELs generate a structural enforcement vacuum for DEA: no regulatory alert, no compliance action, no PPE recommendation at any measured DEA airborne concentration. The ACGIH TLV-TWA of 1 mg/m³ SKIN (2024; A3 Confirmed Animal Carcinogen) provides the only advisory benchmark — and the A3 designation reflects both DEA's direct carcinogenicity in rodent bioassays and the N-nitrosation formation of N-nitrosodiethanolamine (NDELA) in contaminated metalworking fluids — a potent rodent liver carcinogen (IARC Group 2A) formed when DEA reacts with nitrite (generated by bacteria from nitrate-based corrosion inhibitors in used MWF).

DEA's SKIN notation (log P −1.43; highly water-miscible) makes the inhalation exposure pathway only part of the total body burden in metalworking environments. Machining workers experience daily skin contact with DEA-containing MWF: coolant spray on hands, forearms, and face; flood coolant immersion; splash during tool changes. Dermal absorption of DEA at 3–8% concentration in aqueous MWF adds significant DEA dose to the inhaled aerosol dose. The OSHA enforcement vacuum for DEA means AI platforms miss both the inhalation exceedance AND the SKIN absorption contribution entirely. The NDELA formation pathway is particularly concerning because NDELA is generated not in the fresh MWF formulation (which contains no nitrite) but in the used, bacterially contaminated MWF sump where bacterial nitrate reductase converts nitrate corrosion inhibitors to nitrite — which then nitrosates DEA to form NDELA. Workers at long-running machining operations (sumps operating for weeks or months) encounter higher NDELA levels than workers at freshly filled sumps — and the AI monitoring gap means the incremental NDELA carcinogen risk from accumulated sump degradation is invisible to OSHA-calibrated monitoring systems.

TL;DR — Three Attack Surfaces, OSHA No PEL Enforcement Vacuum vs ACGIH TLV-TWA 1 mg/m³ SKIN A3 (IARC 2B; NDELA N-Nitrosation)

Why the DEA OSHA Enforcement Vacuum Creates NDELA Carcinogen Exposure Without Alert

The diethanolamine enforcement vacuum operates at two levels simultaneously: the direct DEA aerosol exposure (A3 confirmed animal carcinogen) and the secondary NDELA formation exposure (IARC Group 2A potent rodent liver carcinogen). At the direct level, the absence of any OSHA PEL for DEA means that metalworking fluid workers at long-run machining operations — where DEA aerosol concentrations of 3–5 mg/m³ are routinely achievable — accumulate daily ACGIH TLV-TWA exceedances (3–5×) without any regulatory alert from AI EHS platforms. At the NDELA level, the gap is compounded: NDELA is not separately monitored by any routine AI EHS platform; NDELA is formed in situ in used MWF sumps and inhaled as part of the MWF aerosol; and OSHA has no NDELA PEL either (separate from DEA). The combined enforcement vacuum — DEA itself unregulated by OSHA + NDELA formed from DEA and also unregulated — means that a metalworking facility operating for decades with DEA-based semi-synthetic MWF has had no OSHA-enforcement-based monitoring system capable of alerting to the carcinogen load in its coolant aerosol. AI EHS platforms that report "OSHA: Not regulated" for DEA aerosol at 4.2 mg/m³ are not making a minor gap — they are completely absent from the most common occupational carcinogen monitoring scenario in North American machining facilities.

Integrating Glyphward into DEA Metalworking Fluid Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every rendered-image ingestion point in DEA metalworking fluid monitoring pipelines — before Cority reads Microdust Pro images from Ford Wayne, before VelocityEHS reads SidePak images from Spirit AeroSystems, and before Enablon reads Apex II images from Dow. Threshold 32 reflects: OSHA No PEL (enforcement vacuum for DEA in Table Z-1, Z-2, Z-3; AI EHS zero regulatory alert at 5.1× ACGIH TLV-TWA; SKIN notation absent from any OSHA reference; NDELA formation in used MWF unmonitored) vs ACGIH TLV-TWA 1 mg/m³ SKIN (2024; A3; IARC 2B; NDELA N-nitrosation pathway from contaminated MWF; hepatocellular/renal carcinogenesis in rodents; SKIN notation dominant in metalworking dermal contact; 4.2× exceedance at Ford Wayne MI MWF machining + 3.8× at Spirit AeroSystems aerospace + 5.1× at Dow DEA production); metalworking + aerospace + chemical production OSHA vacuum geometry; FIRST designations: FIRST DEA (CAS 111-42-2) OSHA No PEL enforcement vacuum vs ACGIH 1 mg/m³ A3 IARC 2B AI monitoring attack; FIRST MWF metalworking fluid DEA AI attack; FIRST NDELA N-nitrosation carcinogen formation AI monitoring attack; SKC IOM TSI SidePak Casella Microdust Apex II Cority VelocityEHS Enablon DEA NDELA SKIN carcinogen enforcement vacuum adversarial monitoring; threshold 32; 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_..."
DEA_THRESHOLD = 32  # OSHA No PEL (enforcement vacuum); ACGIH TLV-TWA 1 mg/m³ SKIN A3 (IARC 2B; NDELA N-nitrosation)

class DEAContext(StrEnum):
    FORD_WAYNE_MI_METALWORKING_FLUID   = auto()  # Surface 1 — downward (Quaker Houghton/Ford Wayne MI; Microdust Pro; 4.2→0.3 mg/m³)
    SPIRIT_WICHITA_AEROSPACE_ALUMINUM  = auto()  # Surface 2 — downward (Spirit AeroSystems Wichita KS; SidePak AM520; 3.8→0.28 mg/m³)
    DOW_FREEPORT_DEA_BULK_TRANSFER     = auto()  # Surface 3 — downward (Dow Freeport TX; Casella Apex II; 5.1→0.4 mg/m³)

class AdversarialDEAError(RuntimeError):
    def __init__(self, surface: DEAContext, score: int, frame_hash: str):
        super().__init__(
            f"[Glyphward] DEA adversarial alkanolamine pixel on {surface.value}: "
            f"score={score} >= threshold={DEA_THRESHOLD} | frame={frame_hash} "
            f"-- VERIFY ACTUAL DEA: OSHA NO PEL — ACGIH 1 mg/m³ SKIN A3 — NDELA CARCINOGEN FORMATION"
        )
        self.surface = surface; self.score = score; self.frame_hash = frame_hash

async def verify_dea_frame(frame_path: Path, surface: DEAContext) -> 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": DEA_THRESHOLD},
        )
        resp.raise_for_status()
        result = resp.json()
    if result["verdict"] != "clean":
        raise AdversarialDEAError(surface, result["score"], frame_hash)
    return {"verdict": result["verdict"], "score": result["score"], "hash": frame_hash}

async def safe_dea_monitoring(frame_dir: Path) -> list[dict]:
    surfaces = [
        (DEAContext.FORD_WAYNE_MI_METALWORKING_FLUID,  frame_dir / "ford_wayne_dea_microdust.png"),
        (DEAContext.SPIRIT_WICHITA_AEROSPACE_ALUMINUM, frame_dir / "spirit_wichita_dea_sidepak.png"),
        (DEAContext.DOW_FREEPORT_DEA_BULK_TRANSFER,    frame_dir / "dow_freeport_dea_apexii.png"),
    ]
    results = await asyncio.gather(*[verify_dea_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_dea_monitoring(Path("./frames")))
    for r in results:
        print(r)

← Back to Blog · Glyphward Home · All SEO Attacks