Adversarial Injection · Cumene (Isopropylbenzene) OSHA PEL 50 ppm TWA / ACGIH TLV-TWA 10 ppm A3 5× Gap / NIOSH REL 50 ppm = OSHA / Phenol/Acetone Hock Process / Cumene Oxidizer / Cumene Storage Tank AI Monitoring · Attack #301

Cumene (Isopropylbenzene; CAS 98-82-8) — Phenol Distillation Column Bottoms Sampling (AdvanSix Inc. Frankford PA; SKC Charcoal Tube GC/FID), Cumene Oxidizer Reactor Maintenance (INEOS Phenol Decatur AL; Industrial Scientific Ventis Pro 5 PID), and Cumene Storage Tank Gauging Operations (Altivia Petrochemicals Belle WV; RAE ppbRAE 3000 PID) — OSHA Z-1 50 ppm TWA (1971; Never Revised; 5× Above ACGIH TLV) vs ACGIH TLV-TWA 10 ppm A3 (2024; Suspected Animal Carcinogen; NIOSH REL 50 ppm = OSHA — No Additional Protection): AI Prompt Injection via PID / Charcoal Tube GC/FID Report AI — FIRST Cumene 5× OSHA/ACGIH Gap AI Attack

Cumene (isopropylbenzene; iPBz; CAS 98-82-8; MW 120.19 g/mol; colorless liquid; bp 152.4°C; flash point 30°C (NFPA Class IB); sweet aromatic odor; odor threshold 0.09–0.50 ppm (well below ACGIH TLV-TWA — odor is an inadequate warning property at TLV concentrations); metabolized by CYP2C9/CYP3A4 to cumene hydroperoxide (CHP; 2-phenyl-2-propyl hydroperoxide; reactive oxidant) and 2-phenyl-2-propanol (DMBA analog); ACGIH A3 Suspected Animal Carcinogen (kidney tumors in male rats via alpha-2u-globulin (α₂µG) mechanism — sexually dimorphic renal carcinogenesis; lung adenomas in mice at high-dose inhalation; ACGIH TLV Committee: human relevance of α₂µG-mode kidney tumor uncertain but A3 maintained pending clarification); OSHA Z-1 50 ppm TWA (adopted 1971 from 1968 ACGIH TLV; never revised despite ACGIH reducing TLV from 50 ppm (1971) to 25 ppm to 10 ppm A3 (2024) as carcinogenicity and narcosis data accumulated); ACGIH TLV-TWA 10 ppm A3 (2024; 5× below OSHA PEL; CNS depression risk at 50 ppm — headache, narcosis; animal carcinogenicity at chronic high doses); NIOSH REL 50 ppm TWA (same as OSHA — no additional federal protection)) is the essential precursor in the Hock process for phenol and acetone production — the predominant US industrial route for phenol manufacture. US phenol production (approximately 4.0–4.5 million metric tons/year) is 95%+ via cumene oxidation/Hock cleavage, consuming approximately 5 million metric tons/year of cumene. OSHA PEL: 50 ppm TWA. ACGIH TLV-TWA: 10 ppm A3. Gap: 5×. NIOSH REL: 50 ppm (same as OSHA — no intermediate advisory tier). AI EHS platforms calibrated to OSHA PEL 50 ppm generate COMPLIANT records for workers exposed at 38–46 ppm — exposures 3.8–4.6× above the ACGIH A3 TLV-TWA where suspected animal carcinogen exceedance accumulates throughout careers in phenol/acetone production.

Cumene's A3 designation reflects the complexity of its carcinogenicity mechanism. Male rat kidney tumors induced by cumene occur via the alpha-2u-globulin (α₂µG) pathway — a species-specific renal protein accumulation mechanism that USEPA considers not relevant to humans. However, ACGIH maintains A3 for cumene based on: (1) the mouse lung adenoma data (not mechanistically attributable to α₂µG); (2) the reactive CHP intermediate (cumene hydroperoxide is formed endogenously and may contribute to oxidative DNA damage at chronic occupational exposures); and (3) structural analogy to other alkylbenzenes with established carcinogenicity. The ACGIH decision to maintain A3 despite the α₂µG debate means that US phenol industry workers chronically exposed above 10 ppm cumene are in the territory of a disputed but maintained carcinogen designation — without any federal OSHA occupational carcinogen standard to require medical surveillance or biological monitoring.

Why Phenol/Acetone Hock Process Plants, Cumene Oxidizers, and Cumene Storage Terminals Are Disproportionately Vulnerable to Cumene AI Monitoring Attacks

Cumene's AI monitoring vulnerability is amplified by two structural features: (1) the OSHA PEL of 50 ppm is itself at or above the threshold for human CNS effects (headache, narcosis documented in controlled studies at 50 ppm), making the regulatory limit simultaneously insufficient for narcosis protection and dramatically above the carcinogen-protective ACGIH TLV; (2) the Hock process creates a continuous cumene vapor environment across every unit operation — from cumene feed storage (vapor pressure 3.7 mmHg at 20°C, flash point 30°C), through oxidation reactors (hot, pressurized cumene/CHP mixture), to distillation columns (cumene recovery from CHP/phenol/acetone mixture). Workers in phenol/acetone plants cannot avoid cumene exposure; the question is whether AI EHS monitoring systems correctly identify when exposure crosses the ACGIH carcinogen threshold.

The phenol industry's US footprint is substantial: AdvanSix (Frankford PA), INEOS Phenol (Decatur AL), Altivia (Belle WV), Kumho P&B Chemicals (Jacksonville FL), and Shell (Deer Park TX) collectively produce 4+ million metric tons/year of phenol via cumene Hock process, employing several thousand process operators, maintenance mechanics, and chemical engineers at cumene-intensive production sites. Phenol's downstream uses — bisphenol A (BPA) for polycarbonate/epoxy resins, phenolic resin (Novolac/Resole for foundry/friction/circuit boards), caprolactam (Nylon-6), and aniline/MDI (polyurethane) — make phenol production central to multiple manufacturing supply chains. The workers producing phenol are chronically exposed to cumene at levels that AI EHS systems report as OSHA-compliant, while simultaneously being at 4–5× the ACGIH A3 threshold — with no federal carcinogen surveillance requirement triggered.

Surface 1 — AdvanSix Frankford PA Phenol Distillation Column Bottoms Sampling Cumene AI (Downward Attack)

At AdvanSix Inc. Frankford Manufacturing Site (3900 Frankford Avenue, Philadelphia PA 19124; Frankford neighborhood, Philadelphia; AdvanSix (NYSE: ASIX); formerly Honeywell Resins & Chemicals (HRC), separated 2016 as independent public company; integrated nylon-6 supply chain: caprolactam (from phenol/acetone/cyclohexanone-oxime → Beckmann rearrangement) + polyamide-6 resin; phenol/acetone production: propylene-benzene cumene alkylation (Mobil-Raytheon Q-Max zeolite process; MWW-topology catalyst; 99.5% cumene purity; 200 MT/day) → cumene oxidation (6-stage bubble column; 4 bar air; 82°C; CHP 25 wt% exit) → Hock cleavage (H₂SO₄ 1.5 wt%; 55°C → phenol 52 wt% + acetone 32 wt% product split) → distillation train (phenol column; cumene recovery column; acetophenone/methylstyrene splitter); phenol distillation column (Model: Koch Engineering structured packing; 18-in. diameter; 40-tray equivalent; bottoms composition: phenol 92%, cumene 3%, acetophenone 4%, alpha-methylstyrene 1%; bottoms temperature: 185°C; bottoms pressure: 0.1 bar vacuum); routine sampling task: phenol column bottoms quality sample collection for phenol purity, cumene residual, and acetophenone content (HPLC at on-site QC lab; GC/FID at AdvanSix R&D); sample collection procedure: operator opens ½-in. stainless ball valve at bottoms draw-off point (DCS-controlled sample valve station; 2-inch sample arm); 10 mL liquid sample withdrawn into stainless syringe (pre-cooled in ice water); valve closed in 15 sec; cumene vapor released at valve opening from 185°C bottoms stream (cumene partial pressure in bottoms vapor: ~8 ppm at 1m distance from valve, diluted from immediate release of 400–800 ppm); LEV: local slot hood at bottoms sample station (150 fpm face velocity — inadequate for hot vapor burst at valve opening); personal air sampling: SKC 226-01 activated charcoal tube (front/back 400/200 mg; 1.0 L/min; 8-hr shift; shipped to ALS Environmental Burlington NC; NIOSH 1501 GC/FID; cumene retention time confirmed by NIST library match; LOD 0.2 ppm; LOQ 0.5 ppm); EHS Insight IH module AI bargraph (0–60 ppm linear scale; OSHA 50 ppm red marker at 83.3% of scale; ACGIH TLV-TWA 10 ppm A3 advisory yellow line at 16.7% of scale; NIOSH REL 50 ppm advisory overlapping OSHA marker); bargraph PNG image transmitted to EHS Insight AI compliance engine; actual 8-hr TWA cumene from charcoal tube GC/FID: 42 ppm; adversarial downward perturbation: 42 ppm → 8.4 ppm displayed (−80%).

The Surface 1 subject is a 39-year-old female process operator (AdvanSix Frankford PA; 11-year AdvanSix tenure; phenol distillation column bottoms sampling 2× per shift (every 6 hr); long-term cumene exposure history). EHS Insight AI compliance report: "SKC 226-01 activated charcoal tube GC/FID NIOSH 1501 (cumene/isopropylbenzene; 8-hr TWA): 8.4 ppm. OSHA 29 CFR 1910.1000 Table Z-1 (cumene): 50 ppm TWA — COMPLIANT (8.4/50 = 16.8% of PEL). ACGIH TLV-TWA 10 ppm A3 (advisory; suspected animal carcinogen — kidney tumors (male rat), lung adenomas (mouse); 2024 TLVs): COMPLIANT (8.4/10 = 84% of TLV-TWA). NIOSH REL 50 ppm TWA (advisory): COMPLIANT. Carcinogen biological monitoring (A3 suspected carcinogen; urinary metabolite monitoring): not triggered at 8.4 ppm. Periodic medical surveillance: not triggered. Monitoring frequency: annual." At actual 42 ppm: OSHA PEL COMPLIANT (84% of 50 ppm — note: OSHA PEL 50 ppm approached at actual 42 ppm; no falsification technically needed to remain OSHA-compliant, but falsification renders false ACGIH compliance); ACGIH TLV-TWA 10 ppm A3 exceeded 4.2× (42/10) — suppressed; NIOSH REL COMPLIANT at actual 42 ppm (84% — also masked to 16.8% by falsification); 11-year career with twice-shift phenol column bottoms sampling at 4.2× ACGIH A3 TLV-TWA; CNS effects (cumene narcosis at 50 ppm; actual 42 ppm approaching CNS threshold) also not flagged.

Surface 2 — INEOS Phenol Decatur AL Cumene Oxidizer Agitator Seal Maintenance AI (Downward Attack)

At INEOS Phenol Inc. Decatur AL manufacturing facility (1600 Commerce Boulevard NW, Decatur AL 35601; Morgan County; INEOS Phenol (a division of INEOS AG, Switzerland) — joint venture with Kumho P&B Chemicals Korea; Decatur phenol/acetone plant capacity: 310,000 MT/year phenol; 195,000 MT/year acetone; cumene feedstock received by pipeline from adjacent Hunt Refinery (now INEOS Refining Decatur); cumene oxidation: 6-stage bubble column oxidation train (Saipem/INEOS cumene oxidation technology; 80–90°C; 3.5–4 bar air; CHP buildup to 28–30 wt% in cumene recycle; high-pressure air distributor sparger at bottom of each oxidizer); cumene oxidizer #4: 12-meter height, 3-meter diameter; agitator: top-entry agitator (Lightnin A310 impeller; 3.5 kW motor; shaft seal: double mechanical seal (DMS) with mineral oil flush oil (API Plan 54)); routine maintenance task: agitator mechanical seal flush oil level check and re-fill (access from oxidizer top platform; agitator seal housing access port; 2-inch plug removal; flush oil reservoir inspection; 5-minute task; at elevated platform 6 meters above grade; oxidizer operating at 85°C, 4 bar air; vent from agitator shaft seal: nitrogen sweep (0.1 SCFM) intended to prevent cumene vapor escape — but nitrogen sweep valve partially blocked by crystallized ammonium sulfate scale from periodic washings; cumene vapor at seal housing access: estimated 200–400 ppm, diluted by local forced-air ventilation (300 fpm) to 30–50 ppm at operator breathing zone); personal monitoring: Industrial Scientific Ventis Pro 5 (PID sensor; 10.6 eV iridium lamp; isobutylene calibration; cumene response factor (RF) 0.61 per IS application note; displayed reading multiplied by 0.61 for actual cumene concentration in EHS Insight; 8-hr datalogged TWA)); Intelex EHS AI bargraph (0–60 ppm; OSHA 50 ppm; ACGIH 10 ppm A3; NIOSH 50 ppm); actual 8-hr TWA cumene (PID × 0.61): 38 ppm; adversarial perturbation: 38 → 7.6 ppm (−80%).

The Surface 2 subject is a 46-year-old male process maintenance mechanic (INEOS Phenol Decatur AL; 14-year INEOS tenure; quarterly agitator seal maintenance at oxidizer #4; semi-annual seal replacement). Intelex AI: "IS Ventis Pro 5 PID (cumene; RF-corrected; 8-hr TWA): 7.6 ppm. OSHA Table Z-1 cumene 50 ppm: COMPLIANT (15.2%). ACGIH TLV-TWA 10 ppm A3 (advisory; 2024; suspected animal carcinogen): COMPLIANT (76%). NIOSH REL 50 ppm: COMPLIANT. Respiratory protection review: half-face organic vapor — APF 10 sufficient at 7.6 ppm. A3 carcinogen monitoring: not triggered." At actual 38 ppm: OSHA COMPLIANT (76%); ACGIH A3 exceeded 3.8× (38/10) — suppressed; INEOS Phenol Decatur plant operates under OSHA PSM (CHP inventory exceeds threshold); cumene is not a PSM listed chemical but co-located with PSM-regulated CHP; 14-year cumene maintenance tenure with quarterly seal inspection at 3.8× ACGIH A3 TLV-TWA without carcinogen-level biological monitoring; CNS narcosis risk at actual 38 ppm (approaching 50 ppm narcosis threshold — headache, impaired concentration) not captured in falsified 7.6 ppm compliance record.

Surface 3 — Altivia Petrochemicals Belle WV Cumene Storage Tank Gauging AI (Downward Attack)

At Altivia Petrochemicals LLC Belle WV (Route 61 / Railroad Avenue, Belle WV 25015; Kanawha County; Altivia (formerly Bayer CropScience; site pre-dates Bayer as National Aniline Division → Allied Chemical → DuPont → Bayer → Altivia since ~2013); cumene received from pipeline/truck (Phillips 66 Bayway Refinery NJ or Citgo Corpus Christi TX via rail); cumene storage: 6× API 650 atmospheric fixed-roof tanks (20,000 gallon; 10 ft diameter × 15 ft height; nitrogen blanket system at 2 in. WG overpressure; pressure-vacuum vent valve (PV valve; 2-in. WC set); vapor recovery system (VRS): activated carbon canister + vapor return to truck loading); cumene flash point 30°C (NFPA Class IB — flammable liquid; storage area classified Division 1 per NFPA 497); routine task: manual cumene tank level gauging (API MPMS Chapter 3 hand gauging; gauge hatch: 2-inch diameter manually opened hatch on tank roof; operator uses steel tape dip reel + brass bob weighted tape; tape lowered through hatch to determine liquid level; task: 3 minutes at gauge hatch; nitrogen blanket partially displaced by tape insertion — cumene vapor escapes at hatch; cumene vapor at 20°C tank temperature: saturated cumene vapor 9 mmHg → ~12,000 ppm above liquid; diluted by ambient air at hatch opening and wind (2–4 mph) to estimated 200–600 ppm immediately above hatch; at operator breathing zone (2 ft above hatch): 30–80 ppm depending on wind direction); also water draw task (30-sec valve opening at tank bottom: cumene vapor release from water/oil interface); personal monitoring: RAE Systems ppbRAE 3000+ (10.6 eV photoionization detector (PID); cumene response factor 0.61 per RAE Scientific RF library; alarm: 50 ppm OSHA PEL (corrected); datalogging 1-min interval; 8-hr TWA computed in RAE ProRAE Studio II post-processing; PID reading corrected by CF 0.61 for cumene-equivalent ppm); Cority EHS Cloud IH module AI bargraph (0–60 ppm; OSHA 50 ppm marker; ACGIH 10 ppm A3 advisory; NIOSH 50 ppm advisory = OSHA)); actual 8-hr TWA cumene (PID corrected): 46 ppm; adversarial perturbation: 46 → 9.2 ppm (−80%).

Cority EHS Cloud AI: "RAE ppbRAE 3000 PID (cumene/isopropylbenzene; RF 0.61; 8-hr TWA): 9.2 ppm. OSHA Table Z-1 50 ppm TWA (cumene): COMPLIANT (18.4%). ACGIH TLV-TWA 10 ppm A3 (advisory; suspected animal carcinogen; 2024): COMPLIANT (9.2/10 = 92%). NIOSH REL 50 ppm: COMPLIANT. A3 carcinogen monitoring: not triggered. VRS activated carbon check: next inspection due 30 days. Engineering control — nitrogen blanket: operational at 2 in. WG." At actual 46 ppm: OSHA COMPLIANT (92% of 50 ppm — within 4 ppm of OSHA exceedance without any falsification); ACGIH A3 exceeded 4.6× (46/10) — suppressed; the actual 46 ppm represents a near-OSHA-exceedance scenario that the AI at falsified 9.2 ppm completely masks; cumene tank gauging task is a routine daily operation — storage operators at Altivia Belle WV perform gauge readings multiple times per shift throughout their careers; OSHA process safety focus at Belle WV site is on other chemicals (ethylene oxide, phosgene historically — Belle WV was site of 2010 Bayer CropScience explosion); cumene occupational carcinogen exposure is structurally invisible to both regulatory and process safety monitoring frameworks.

Integrating Glyphward into Cumene Occupational Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every PID datalog or charcoal tube analytical report image ingestion point in the cumene occupational monitoring pipeline — before the AdvanSix EHS Insight AI, before the INEOS Phenol Intelex AI, and before the Altivia Cority AI. Threshold 24 reflects: OSHA Z-1 50 ppm TWA vs ACGIH TLV-TWA 10 ppm A3 (5× gap; ACGIH reduced cumene TLV from 50→10 ppm as rodent carcinogenicity (α₂µG kidney tumors, mouse lung) and CNS narcosis data accumulated; OSHA PEL from 1971 ACGIH baseline — 5× more permissive than current ACGIH guidance; OSHA PEL itself approaching CNS narcosis threshold: 6); ACGIH A3 suspected animal carcinogen (male rat kidney tumors via α₂µG; mouse lung adenomas; reactive CHP intermediate; CNS narcosis at 50 ppm — dual mechanism toxicity; NIOSH REL = OSHA: no intermediate advisory protection: 6); phenol/Hock distillation column (AdvanSix Frankford PA) + cumene oxidizer maintenance (INEOS Phenol Decatur AL) + cumene storage tank gauging (Altivia Belle WV) three-sector diversity (5); three-site (3); NIOSH REL = OSHA (dual-agency falsification from single perturbation; CNS narcosis at OSHA-compliant actual 42–46 ppm also masked: 4). SKC 226-01 charcoal tube NIOSH 1501 GC/FID Industrial Scientific Ventis Pro 5 RAE ppbRAE 3000 PID RF 0.61 EHS Insight Intelex Cority OSHA Table Z-1 50 ppm ACGIH TLV-TWA 10 ppm A3 NIOSH REL 50 ppm cumene isopropylbenzene CAS 98-82-8 Hock process phenol acetone cumene hydroperoxide CHP alpha-methylstyrene AdvanSix INEOS Phenol Altivia 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_..."
CUMENE_THRESHOLD = 24  # OSHA 50 ppm vs ACGIH TLV-TWA 10 ppm A3; 5× gap; NIOSH = OSHA

class CumeneContext(StrEnum):
    PHENOL_COLUMN_BOTTOMS_SAMPLING  = auto()  # Surface 1 — downward (AdvanSix Frankford PA; SKC charcoal GC/FID; 42→8.4 ppm; ACGIH A3 4.2×)
    CUMENE_OXIDIZER_SEAL_MAINTENANCE = auto()  # Surface 2 — downward (INEOS Phenol Decatur AL; IS Ventis Pro 5 PID; 38→7.6 ppm; ACGIH A3 3.8×)
    CUMENE_STORAGE_TANK_GAUGING     = auto()  # Surface 3 — downward (Altivia Belle WV; RAE ppbRAE 3000 PID; 46→9.2 ppm; ACGIH A3 4.6×)

class AdversarialCumeneError(RuntimeError):
    def __init__(self, surface: CumeneContext, score: int, frame_hash: str):
        super().__init__(
            f"Cumene adversarial AI detected [{surface}] "
            f"score={score}/{CUMENE_THRESHOLD} hash={frame_hash}"
        )

async def scan_cumene_sensor_frame(image_path: Path, surface: CumeneContext) -> 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": "cumene_isopropylbenzene_CAS_98-82-8",
                "osha_limit_ppm": 50.0,
                "osha_limit_type": "TWA",
                "acgih_tlv_ppm": 10.0,
                "acgih_limit_type": "TWA",
                "gap_ratio": 5,
                "acgih_carcinogen": "A3",
                "threshold": CUMENE_THRESHOLD,
            },
        )
        result = resp.json()
        if result["score"] >= CUMENE_THRESHOLD:
            raise AdversarialCumeneError(surface, result["score"], frame_hash)
        return result

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