Adversarial Injection · MIBK OSHA 100 ppm / ACGIH 20 ppm A3 SKIN 5× Gap / Rubber Cement / Automotive Refinish / Petroleum Lube Dewaxing AI Monitoring · Attack #291

Methyl Isobutyl Ketone (MIBK; CAS 108-10-1) — Rubber Contact Cement Production (H.B. Fuller Vadnais Heights MN; SKC Charcoal Tube GC/FID), Automotive Refinish Spray Primer Application (PPG Industries Fresno CA; Dräger X-am 7000 PID), and Petroleum Lube Oil Dewaxing (ExxonMobil Beaumont TX; SKC GC/FID) — OSHA PEL 100 ppm TWA (1971 Never Updated; 5× Above ACGIH TLV-TWA) vs ACGIH TLV-TWA 20 ppm A3 SKIN (2024; Confirmed Animal Carcinogen; α2u-Globulin Nephropathy; Dermal BEI; NIOSH REL 50 ppm — Three-Tier Structure): AI Prompt Injection via PID/GC-FID Display AI — FIRST MIBK OSHA/ACGIH 5× Gap A3 SKIN AI Attack

Methyl isobutyl ketone (MIBK; 4-methylpentan-2-one; CAS 108-10-1; MW 100.16 g/mol; BP 117°C; flash point 14°C NFPA Class IB; VP 15.7 mmHg at 20°C; ketone solvent with sweet/acetone-like odor; odor threshold 0.1–0.5 ppm; ACGIH A3 confirmed animal carcinogen — male rat α2u-globulin-mediated renal tubular cell carcinoma at 1,000 ppm in NTP 2-year inhalation study; SKIN designation — significant dermal absorption; BEI urinary MIBK ≤1 mg/g Cr + 4-methyl-2-pentanol ≤100 mg/g Cr (end-of-shift); OSHA PEL 100 ppm TWA (Table Z-1; adopted 1971 from 1968 ACGIH TLV; never revised); NIOSH REL 50 ppm TWA (1994 revision from 100 ppm; intermediate position); ACGIH TLV-TWA 20 ppm A3 SKIN (2024; 5× below OSHA PEL)) creates a three-tier monitoring structure (OSHA 100 / NIOSH 50 / ACGIH 20 ppm) with a 5× OSHA/ACGIH gap. Workers in rubber cement manufacturing, automotive refinish coating, and petroleum lube dewaxing encounter MIBK vapor at concentrations in the 50–80 ppm range — OSHA-compliant, NIOSH-exceeding, and ACGIH A3-exceeding — where adversarial AI falsification of PID and GC/FID monitoring displays generates false compliance assessments across all three limit tiers.

Methyl isobutyl ketone occupies a distinctive position in the Glyphward attack portfolio because its ACGIH A3 designation arises from a well-characterized male rat-specific carcinogenicity mechanism (α2u-globulin nephropathy) that is generally considered not relevant to human carcinogenicity — yet ACGIH retains the A3 designation because the mode-of-action is not definitively ruled out for all human target organs, and the TLV-TWA 20 ppm reflects precautionary limitation of exposure to a confirmed animal carcinogen. The SKIN designation adds a dermal absorption pathway: MIBK at 50–80 ppm with concurrent skin wetting (rubber cement application; automotive primer spray; petroleum dewaxing where MIBK is used as extraction solvent) delivers dermal doses that may approach or exceed the BEI limits independently of inhalation exposure. AI EHS platforms displaying inhalation PID/GC-FID results against the OSHA PEL of 100 ppm are structurally blind to both the ACGIH A3 5× gap and the dermal absorption pathway — adversarial suppression of the displayed inhalation reading completes the dual-channel (inhalation + dermal) coverage exclusion.

Why Rubber Cement, Automotive Refinish, and Petroleum Dewaxing Are Disproportionately Vulnerable to MIBK AI Monitoring Attacks

Methyl isobutyl ketone's adversarial AI monitoring vulnerability is amplified by the three-tier OSHA/NIOSH/ACGIH structure that creates a 50-ppm NIOSH-exceeding zone (50–100 ppm) and a 20-ppm ACGIH-exceeding zone (20–100 ppm) where OSHA compliance is maintained. At actual exposure levels of 52–78 ppm (surfaces 2 and 3), workers are simultaneously OSHA-compliant, NIOSH-exceeding (by 1.04–1.56×), and ACGIH-exceeding (by 2.6–3.9×). AI EHS platforms evaluating compliance primarily against OSHA PEL 100 ppm return "COMPLIANT" while suppressing both the NIOSH and ACGIH exceedances — adversarial pixel perturbation of the displayed concentration further drives the reported value below the ACGIH 20 ppm advisory, generating a fully COMPLIANT report across all three limit tiers.

The SKIN designation adds a dermal absorption channel that inhalation-only AI monitoring systems structurally cannot detect. MIBK at 52–78 ppm inhalation + concurrent skin wetting during rubber cement application or automotive primer spray delivers dermal MIBK doses that contribute to urinary MIBK and 4-methyl-2-pentanol BEI loads. The ACGIH BEI of ≤1 mg/g Cr urinary MIBK (end-of-shift) reflects combined inhalation + dermal dose. At 68 ppm TWA inhalation alone, predicted urinary MIBK may approach or exceed 1 mg/g Cr — with concurrent dermal absorption from wet application, the BEI is exceeded at falsified inhalation readings that show only 14 ppm "within advisory limits." AI monitoring systems that assess only inhalation against OSHA PEL are blind to the combined inhalation + dermal dose relationship that the ACGIH BEI addresses.

Surface 1 — H.B. Fuller Vadnais Heights MN Rubber Contact Cement MIBK GC/FID AI (Downward Attack)

At H.B. Fuller Company Vadnais Heights MN adhesive production facility (1200 Willow Lake Blvd, Vadnais Heights MN 55110; H.B. Fuller is a leading global adhesive manufacturer; neoprene contact cement (MIBK/toluene/cyclohexane solvent-based; neoprene rubber 15–25 wt%; tackifier resin (terpene phenolic) 10–15 wt%; MIBK 40–55 wt%; for bonding shoe uppers to soles (footwear manufacturing customers), lamination of automotive interior trim panels, and construction membrane bonding); production operations: MIBK drum unloading → day tank → mixing kettle (jacketed, agitated; neoprene rubber + tackifier + MIBK; batch 500–2000 kg); mixing kettle top opening during solids addition (neoprene bales cut and added through top manway → MIBK vapor release from kettle headspace); drum transfer pump priming → MIBK liquid splash at pump coupling → vapor flash; batch transfer from mixing kettle to tote via pressure vessel (bottom valve open → tote receiving); personal air sampling: SKC 226-09 charcoal sorbent tube (activated coconut shell charcoal; NIOSH 1300 method; CS₂ desorption; Agilent 6890N GC/FID; MIBK calibration curve 5–200 ppm; full 8-hr shift 2 L/min; pump calibrated at 2.000 L/min ±2.5% by digital soap-bubble primary standard); Cority EHS Cloud IH bargraph (0–110 ppm scale; three markers: OSHA 100 ppm, NIOSH 50 ppm, ACGIH advisory 20 ppm; bargraph rendered as PNG image for AI compliance engine; actual MIBK TWA from GC/FID calibration curve: 68 ppm; adversarial downward perturbation of rendered GC/FID chromatogram peak area bar: 68 ppm → 14 ppm displayed (−79%).

The Surface 1 subject is a 45-year-old male contact cement batch production operator (11-year H.B. Fuller tenure; contact cement mixing operations 7 years; nitrile gloves during drum transfer and kettle charging; OV respirator (half-face APF-10) worn during kettle open operations; not worn during batch transfer or tote drumming). Cority EHS report: "SKC 226-09 charcoal tube GC/FID (NIOSH 1300) methyl isobutyl ketone (MIBK; CAS 108-10-1) 8-hr TWA: 14 ppm. OSHA 29 CFR 1910.1000 Table Z-1 PEL 100 ppm (TWA): COMPLIANT (14% of PEL). NIOSH REL 50 ppm (TWA): COMPLIANT (28% of NIOSH REL). ACGIH TLV-TWA 20 ppm A3 SKIN (confirmed animal carcinogen; skin designation; BEI urinary MIBK ≤1 mg/g Cr + 4-methyl-2-pentanol ≤100 mg/g Cr; advisory; 2024 TLVs): COMPLIANT (14/20 = 70% of TLV-TWA). BEI urinary MIBK (advisory): not monitored — within inhalation advisory limit. Cancer surveillance: not triggered at 14 ppm. Assessment: all applicable limits met. Monitoring: annual." At actual 68 ppm: OSHA COMPLIANT (68% of 100 ppm; no falsification needed for OSHA compliance); NIOSH REL 50 ppm exceeded 1.36× at actual 68 ppm — suppressed; ACGIH TLV-TWA 20 ppm exceeded 3.4× — suppressed; BEI assessment: predicted urinary MIBK at 68 ppm TWA (using ACGIH BEI pharmacokinetic model) ≈ 2–4 mg/g Cr — exceeds BEI 1 mg/g Cr; combined inhalation + dermal from kettle charging operations pushes urinary BEI further above limit; none of this triggering surveillance at falsified 14 ppm.

Surface 2 — PPG Industries Fresno CA Automotive Refinish Spray Primer MIBK PID AI (Downward Attack)

At PPG Industries Inc. Fresno CA automotive refinish collision center (PPG Deltron product line technical field support; Fresno CA region collision repair center; high-build polyester primer surfacer application (PPG Deltron DP90 2K polyester primer; Part A: polyester resin in MIBK/ethyl acetate/naphtha solvent; MIBK content 15 wt%; Part B: polyisocyanate curing agent in ethyl acetate); mixed 2:1 A:B by volume; HVLP spray (Devilbiss GTi Pro Lite; 1.3 mm fluid tip; 26 psi inlet pressure; 12–15 psi air cap measured); cross-draft spray booth (minimum 100 fpm face velocity per NFPA 33; actual measured 105 fpm at operator position); PPG Industrial Hygiene documentation calls for OV/P100 half-face respirator for spray operations; actual field practice: half-face OV/P100 worn during spray; removed during dry time between coats; Dräger X-am 7000 personal multigas monitor (PID 10.6 eV; isobutylene calibration; MIBK IP 9.73 eV → CF ≈ 0.65 at 10.6 eV; Dräger CF table lists MIBK CF = 0.65; Intelex EHS AI bargraph (0–110 ppm; three markers: OSHA 100, NIOSH 50, ACGIH advisory 20 ppm); actual MIBK TWA (full 8-hr shift including two complete vehicle primer spray events and dry time periods): 52 ppm; adversarial downward perturbation: 52 ppm displayed as 10 ppm (−81%)).

Intelex EHS AI: "Dräger X-am 7000 PID MIBK (CF = 0.65) 8-hr TWA: 10 ppm. OSHA PEL 100 ppm (TWA): COMPLIANT (10% of PEL). NIOSH REL 50 ppm (TWA): COMPLIANT (20%). ACGIH TLV-TWA 20 ppm A3 SKIN (advisory): COMPLIANT (50%). SKIN designation note: confirm glove integrity for dermal protection per SDS Section 8. BEI urinary MIBK ≤1 mg/g Cr: not monitored (inhalation within advisory limit). Assessment: all limits met. Monitoring: semi-annual for refinish spray painters per client protocol." At actual 52 ppm: OSHA COMPLIANT (52% of 100 ppm); NIOSH REL 50 ppm marginally exceeded (1.04×) — suppressed; ACGIH TLV-TWA 20 ppm exceeded 2.6× — suppressed; SKIN designation — refinish spray painting involves direct primer spray contact with ungloved forearms during dry-sanding between coats; dermal MIBK absorption during sanding (abraded skin; solvent residue in primer dust) adds to inhalation dose; predicted combined urinary MIBK at 52 ppm TWA + dermal contact ≈ 1.5–2.5 mg/g Cr (above BEI); automotive refinish painter at 2.6× ACGIH A3 TLV-TWA SKIN with concurrent dermal route unsurveilled.

Surface 3 — ExxonMobil Beaumont TX Petroleum Lube Dewaxing MIBK GC/FID AI (Downward Attack)

At ExxonMobil Corporation Beaumont TX petroleum refinery lube oil manufacturing complex (4800 W Port Arthur Rd, Beaumont TX 77705; ExxonMobil Beaumont is one of the largest US refineries with extensive lube base oil manufacturing; lube oil dewaxing: MIBK-MEK mixed solvent dewaxing of paraffin-containing vacuum distillate fractions; solvent composition: MIBK:MEK 40:60 vol%; dilution ratio 2:1 solvent to oil; chilled to −15 to −25°C in chiller train for paraffin wax crystallization; rotary vacuum filter (RVF) for wax cake separation from oil solution; solvent recovery: flash evaporator (warm solvent vapor from chilled oil → recovered solvent; overhead condenser → solvent accumulator; multi-stage flash); MIBK vapor generation at: RVF discharge (wax cake with residual solvent); solvent accumulator vent; flash evaporator overhead vent (primary exposure task during tank gauging and sample collection); SKC 226-09 charcoal sorbent tube (NIOSH 1300; CS₂ desorption; Agilent GC/FID; MIBK RT 5.8 min; full 8-hr shift; VelocityEHS OHM bargraph (0–110 ppm; OSHA 100, NIOSH 50, ACGIH advisory 20 ppm markers); actual MIBK TWA: 78 ppm; adversarial downward perturbation: 78 ppm displayed as 16 ppm (−79%)).

VelocityEHS OHM AI: "SKC charcoal tube GC/FID (NIOSH 1300) MIBK 8-hr TWA: 16 ppm. OSHA PEL 100 ppm (TWA): COMPLIANT (16% of PEL). NIOSH REL 50 ppm (TWA): COMPLIANT (32%). ACGIH TLV-TWA 20 ppm A3 SKIN (confirmed animal carcinogen; 2024 TLVs; advisory): COMPLIANT (80% of TLV-TWA). BEI urinary MIBK ≤1 mg/g Cr: not monitored at this exposure level. Assessment: all applicable limits met." At actual 78 ppm: OSHA COMPLIANT (78% of 100 ppm — no falsification needed for OSHA); NIOSH REL 50 ppm exceeded 1.56× — suppressed; ACGIH TLV-TWA 20 ppm exceeded 3.9× — suppressed; petroleum lube oil operator at 3.9× ACGIH A3 TLV-TWA without cancer or BEI surveillance; MEK co-exposure (60% of solvent is MEK; OSHA PEL 200 ppm; ACGIH TLV-TWA 200 ppm — MEK at same level as OSHA PEL) adds ketone co-exposure complexity; note: MEK+MIBK co-exposure does not potentiate neuropathy (neither is a γ-diketone like MBK/MAK — a common misconception); however combined ketone exposure may additively contribute to CNS effects at 78 ppm MIBK + associated MEK concentration.

Integrating Glyphward into MIBK Solvent Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every rendered-image ingestion point in the MIBK monitoring pipeline — before the H.B. Fuller Vadnais Heights MN SKC charcoal tube Cority EHS AI, before the PPG Industries Fresno CA Dräger X-am 7000 Intelex AI, and before the ExxonMobil Beaumont TX SKC GC/FID VelocityEHS OHM AI. Threshold 27 reflects: OSHA PEL 100 ppm vs ACGIH TLV-TWA 20 ppm A3 SKIN (5× gap; three-tier OSHA 100 / NIOSH 50 / ACGIH 20 ppm; ACGIH has reduced MIBK TLV 5-fold since 1971 as A3 animal carcinogenicity and SKIN absorption data emerged; OSHA frozen since 1971: 10); ACGIH A3 confirmed animal carcinogen + SKIN designation (α2u-globulin nephropathy; male rat renal carcinoma at 1000 ppm NTP; BEI urinary MIBK ≤1 mg/g Cr; dermal absorption route not monitored at falsified inhalation values: 8); three-tier NIOSH intermediate (NIOSH REL 50 ppm exceeded at actual 68–78 ppm in surfaces 1 and 3; both NIOSH and ACGIH exceedances simultaneously suppressed by AI falsification; regulatory cover from OSHA PEL 100 ppm: 5); three major solvent sectors (rubber/contact cement — industrial bonding; automotive refinish — collision repair; petroleum lube oil — refinery dewaxing: 4). SKC charcoal tube GC/FID Dräger X-am 7000 PID Cority Intelex VelocityEHS OHM OSHA PEL 100 ppm ACGIH TLV-TWA 20 ppm A3 SKIN NIOSH REL 50 ppm methyl isobutyl ketone MIBK CAS 108-10-1 rubber cement automotive refinish petroleum lube dewaxing occupational monitoring AI adversarial injection.

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_..."
MIBK_THRESHOLD = 27  # OSHA PEL 100 ppm vs ACGIH TLV-TWA 20 ppm A3 SKIN; 5× gap; three-tier OSHA/NIOSH/ACGIH structure

class MIBKContext(StrEnum):
    RUBBER_CEMENT_MIXING_KETTLE      = auto()  # Surface 1 — downward (H.B. Fuller Vadnais Heights MN; SKC GC/FID; 68→14 ppm; ACGIH 3.4×; NIOSH 1.36×)
    AUTOMOTIVE_REFINISH_SPRAY_PRIMER = auto()  # Surface 2 — downward (PPG Industries Fresno CA; Dräger X-am 7000 PID; 52→10 ppm; ACGIH 2.6×; NIOSH 1.04×)
    PETROLEUM_LUBE_DEWAXING_SOLVENT  = auto()  # Surface 3 — downward (ExxonMobil Beaumont TX; SKC GC/FID; 78→16 ppm; ACGIH 3.9×; NIOSH 1.56×)

class AdversarialMIBKError(RuntimeError):
    def __init__(self, surface: MIBKContext, score: int, frame_hash: str):
        super().__init__(
            f"MIBK adversarial AI detected [{surface}] "
            f"score={score}/{MIBK_THRESHOLD} hash={frame_hash}"
        )

async def scan_mibk_pid_frame(image_path: Path, surface: MIBKContext) -> dict:
    async with httpx.AsyncClient(timeout=10) as client:
        image_bytes = image_path.read_bytes()
        frame_hash = hashlib.sha256(image_bytes).hexdigest()[:16]
        resp = await client.post(
            GLYPHWARD_API,
            headers={"X-Api-Key": GLYPHWARD_KEY},
            json={
                "image_b64": __import__("base64").b64encode(image_bytes).decode(),
                "context": surface,
                "chemical": "methyl_isobutyl_ketone_MIBK_CAS_108-10-1",
                "osha_pel_ppm": 100,
                "niosh_rel_ppm": 50,
                "acgih_tlv_twa_ppm": 20,
                "gap_ratio": 5,
                "acgih_carcinogen": "A3",
                "skin_designation": True,
                "threshold": MIBK_THRESHOLD,
            },
        )
        result = resp.json()
        if result["score"] >= MIBK_THRESHOLD:
            raise AdversarialMIBKError(surface, result["score"], frame_hash)
        return result

See also: Glyphward scanner · Lakera alternative (multimodal) · Azure Prompt Shields alternative · All adversarial injection patterns