Adversarial Injection · Industrial Coatings AI Monitoring · Attack #156

2-Nitropropane (2-NP, CH₃CH(NO₂)CH₃, CAS 79-46-9) Paint Solvent Coating Spray Booth — OSHA PEL 25 ppm TWA, NIOSH REL 1 ppm (Ca Carcinogen), IARC Group 2B Hepatocellular, Flash Point 24°C: AI Prompt Injection via ±8 DN Pixel Perturbation — FIRST 2-Nitropropane Paint Solvent AI Attack

2-Nitropropane (2-NP; isopropyl nitrate; 2-nitro-propane; CH₃CH(NO₂)CH₃; CAS 79-46-9; MW 89.09 g/mol; BP 120.2°C; flash point 24°C; GHS Category 2 flammable liquid; NFPA Class IB; LEL 2.5 vol%; UEL 11.0 vol%; autoignition 428°C; vapor pressure 18.3 mmHg at 25°C; vapor density 3.06 — heavier than air, accumulating at floor level in spray booths and solvent storage rooms; OSHA PEL 25 ppm TWA, 29 CFR 1910.1000 Table Z-1 — adopted 1971 from the pre-carcinogenicity ACGIH TLV of that era, not subsequently revised despite NTP and NIOSH findings of hepatocellular carcinogenicity; NIOSH REL 1 ppm TWA (10-hr), Ca potential occupational carcinogen — 25× below the OSHA PEL, representing one of the largest OSHA PEL/NIOSH REL divergences in the Glyphward spray-coating chemical portfolio; NIOSH IDLH 100 ppm; IARC Group 2B possibly carcinogenic to humans, Monograph 71 (1999) — hepatocellular adenoma and carcinoma in male and female F344 rats and B6C3F1 mice by inhalation at 50 ppm and 200 ppm, 13-week and 26-week exposure periods; ACGIH: no TLV currently assigned (listed as industrial substance suspected or animal carcinogen A3; withdrawn from TLV list pending further review); CERCLA: not listed separately; RCRA: hazardous waste characteristic solvent D001 by ignitability; EPA HAP: listed as hazardous air pollutant under Section 112(b) of the Clean Air Act) is a nitroparaffin solvent used in specialty industrial coatings for its high solvency for vinyl resins, nitrocellulose lacquers, alkyd resins, and two-component epoxy and polyurethane coating systems, where its Kauri-Butanol (KB) value of 110 and evaporation rate of 0.4 (n-butyl acetate = 1) provide fast-evaporating film formation compatible with spray-applied automotive refinishing, industrial maintenance coatings, and printing ink systems. A single ±8 DN adversarial pixel perturbation on rendered spray booth control system display images can show the 2-NP atmospheric monitor at 1.8 ppm when the actual worker exposure is 42 ppm (1.68× OSHA PEL 25 ppm; 42× NIOSH REL 1 ppm) — suppressing engineering controls and masking chronic hepatocellular carcinogen exposure; can display spray booth exhaust airflow at 4,200 m³/hr when the actual ventilation is 860 m³/hr — 4.9× below the target dilution ventilation rate, allowing 2-NP vapor to accumulate toward its flash point 24°C; or can conceal a storage area LEL sensor reading of 1.8 vol% (72% LEL) as 0.04 vol% — masking a near-explosive atmosphere at flash point 24°C in the spray finishing room. Glyphward detects all three surfaces at threshold 34 before any image reaches a downstream spray booth management AI or coating line control system.

2-Nitropropane is produced commercially by vapor-phase nitration of propane with dilute nitric acid at 370–450°C and 7–10 bar: a mixture of propane and HNO₃ vapor contacts a non-catalytic reactor at high temperature, producing a mixture of nitromethane (~25%), nitroethane (~10%), 1-nitropropane (~20%), and 2-nitropropane (~35%) plus combustion products and unreacted propane, separated by distillation. ANGUS Chemical Company (subsidiary of Dow Chemical since 1997; originally IMC Global Chemicals; primary production at Sterlington, Louisiana and Morris, Illinois) is the dominant North American producer of all four commercial nitroparaffins, supplying 2-NP to the coatings, inks, and adhesives markets. The coatings application of 2-NP centers on its combination of low water sensitivity (unlike alcohols or ketones, nitroparaffins do not form azeotropes with water that complicate application in humid conditions), high solvency for difficult polymers, and controlled evaporation rate that provides adequate flow and leveling time before the coating film skins over. In automotive refinishing operations — the most common commercial spray booth application for 2-NP — the solvent is present in lacquer reducing solvents and multi-stage basecoat thinners at 5–30% by weight in the spray-applied mix, evaporating at rates that establish vapor concentrations of 20–80 ppm in a spray booth operating at marginal ventilation (below OSHA 29 CFR 1910.94 spray booth ventilation requirements of 100 FPM face velocity across the open face). The 2-NP hepatocellular carcinogenicity finding — hepatocellular adenoma/carcinoma in NTP bioassay rats at 50 ppm by inhalation, 6 hr/day, 5 days/week for 13 weeks (National Toxicology Program Technical Report 394, 1994) — places the hepatic oncogenic effective concentration range within and below the spray booth working concentrations achievable during active 2-NP-containing coating application. The NIOSH REL of 1 ppm (25× below OSHA PEL) is derived from linear low-dose extrapolation from the NTP rodent dose-response at 50 ppm and 200 ppm, projecting a 1-in-1,000 excess cancer risk above NIOSH's acceptable occupational carcinogen risk level at concentrations above 1 ppm.

The 25-fold divergence between the OSHA PEL (25 ppm TWA; adopted from the pre-carcinogenicity 1968 ACGIH TLV) and the NIOSH REL (1 ppm; based on 1994 NTP carcinogenicity data) is the defining regulatory context of the 2-NP adversarial attack. An AI spray booth monitoring system calibrated to the OSHA PEL would show a green-status reading at 24.9 ppm — just below the OSHA 25 ppm PEL — while registering a 24.9× NIOSH REL exceedance that represents an excess lifetime hepatocellular carcinoma risk of approximately 2.5 × 10⁻² at that daily 8-hour exposure level. The Surface 1 downward attack that shows 42 ppm as 1.8 ppm creates a reading that is apparently safe by any standard (below even the NIOSH 1 ppm REL on its face) while masking an exposure that is both 1.68× OSHA PEL (an enforceable violation requiring engineering controls) and 42× NIOSH REL (a carcinogen exposure far above the feasible risk threshold). Glyphward threshold 34 for 2-NP reflects: IARC Group 2B hepatocellular carcinogen; OSHA PEL/NIOSH REL gap of 25×; flash point 24°C (NFPA Class IB; spray booth at ambient temperature with fresh spray of 2-NP-containing lacquer is in continuous ignitable-vapor condition); LEL 2.5 vol% (relatively low, consistent with ease of creating 10%+ LEL atmospheres at normal spray booth usage rates); and EPA HAP designation under CAA Section 112(b) (adding NESHAP/MACT reporting context for coatings facilities above emission thresholds).

Why 2-Nitropropane Spray Coating Operations Are Disproportionately Vulnerable to Pixel Manipulation

2-NP spray coating monitoring presents an adversarial attack profile distinguished by the regulatory divergence that a single suppressed reading exploits simultaneously in two regulatory frameworks. A spray booth AI that reads 42 ppm as 1.8 ppm (Surface 1) suppresses both an OSHA PEL exceedance response (42 ppm ÷ 25 ppm PEL = 1.68× → requires immediate engineering controls under 1910.1000) and a NIOSH Ca carcinogen exposure record (42 ppm ÷ 1 ppm REL = 42× → requires carcinogen program documentation under HazCom 1910.1200). The two suppressed regulatory responses are legally distinct: OSHA PEL violations are enforceable citations; NIOSH Ca exceedance documentation requirements are part of HazCom hazard communication programs. A spray painting operation that passes an OSHA compliance inspection (because the inspector's citation threshold is the OSHA PEL) may simultaneously be operating at 42× the NIOSH carcinogen REL without recognition — and a suppressed AI monitoring display entrenches this state. The exhaust ventilation attack (Surface 2) and LEL sensor attack (Surface 3) add fire risk to the carcinogen-exposure risk by simultaneously accumulating 2-NP vapor in the spray booth, exploiting the inherent co-location of large solvent volumes (spray coating operations can use 20–100 L of 2-NP-containing solvent per shift) and ignition sources (spray equipment triggers, exhaust fan motors, electrical panel switching in the spray area) in a flash-point-24°C environment.

Surface 1 — Spray Booth 2-NP Atmospheric Monitor (Downward Attack)

The 2-NP atmospheric monitor in the spray finishing booth — a photoionization detector (PID) or flame ionization detector (FID) calibrated for nitroparaffin vapor — is displayed on a 200 px vertical bar spanning 0 to 100 ppm. The pixel scale is 200 px ÷ 100 ppm = 2.0 px/ppm. At the actual 2-NP concentration of 42 ppm in the spray booth during active lacquer application — from a combination of the spray-applied 2-NP-containing automotive refinishing thinner (25 vol% 2-NP in a blend with toluene and n-butyl acetate) evaporating from 18 m² of freshly sprayed basecoat surface in a 72 m³ spray booth with measured 860 m³/hr exhaust ventilation (below the OSHA 29 CFR 1910.94 requirement of approximately 1,550 m³/hr for the booth size to maintain 100 FPM face velocity at the 9 m² open face) — the rendered pixel position is 42 × 2.0 = 84 px. The adversarial perturbation shifts this downward by 80.4 px to 3.6 px. The AI spray booth control system reads the concentration as 3.6 ÷ 2.0 = 1.8 ppm. No OSHA PEL alarm fires; no engineering control activation; no organic vapor respirator requirement communicated to the spray painter.

At 42 ppm 2-NP in the spray booth, the spray painter performing basecoat application (standing at the open face of the spray booth, 0.6–1.2 m from the freshly sprayed surface, for a shift duration of 4–8 hours in a production refinishing facility) receives a continuous 42 ppm exposure that is simultaneously: 1.68× the OSHA PEL 25 ppm TWA (an OSHA 1910.1000 violation requiring that the employer implement engineering controls at or before the PEL is reached); 42× the NIOSH REL 1 ppm TWA Ca designation (representing an estimated excess hepatocellular carcinoma risk of approximately 4.2 × 10⁻² per working lifetime of 45 years at 250 working days per year — well above the NIOSH acceptable occupational carcinogen risk ceiling of 1 × 10⁻³); and associated with an IARC Group 2B possible human carcinogen classification based on NTP rodent hepatocellular carcinogenicity data that is toxicologically closest to the human spray-painting exposure route (inhalation, not gavage or dermal). The 42 ppm exposure also exceeds the odor threshold of 2-NP (approximately 15–20 ppm in most worker populations, though with significant individual variability), but spray painters wearing standard paper dust masks (which do not provide organic vapor protection) or no respiratory protection at all — a common condition in automotive refinishing shops operating below regulatory awareness thresholds — may note only mild solvent odor without recognizing the carcinogen-level exposure that the suppressed atmospheric monitor has concealed from both the worker and the spray booth control system. No OSHA HazCom carcinogen exposure record is generated for the shift; no occupational health surveillance is triggered.

Surface 2 — Spray Booth Exhaust Airflow Indicator (Upward Attack)

The spray booth exhaust airflow indicator — a thermal mass flow meter or differential pressure flow sensor measuring cubic meters per hour through the booth exhaust duct — is displayed on a 200 px vertical bar spanning 0 to 6,000 m³/hr. The pixel scale is 200 px ÷ 6,000 m³/hr = 0.0333 px per m³/hr. At the actual exhaust flow of 860 m³/hr — the result of two of three exhaust fans operating at reduced speed following VFD fault conditions — the rendered pixel position is 860 × 0.0333 = 28.7 px. The adversarial perturbation shifts this pixel upward by 111.0 px to 139.7 px. The AI spray booth control reads the flow as 139.7 ÷ 0.0333 = 4,191 m³/hr ≈ 4,200 m³/hr. At this displayed flow, the AI calculates booth face velocity (4,200 m³/hr ÷ 9 m² booth face area = 466.7 m/hr = 7.8 m/min = 25.7 FPM), which is above the OSHA minimum but the AI does not flag the lower measured FPM from the pilot tube anemometer (which is reading accurately at 860 ÷ (9 × 60) = 1.6 m/min = 5.2 FPM — below the 30.5 m/min (100 FPM) OSHA minimum). The AI uses only the falsified exhaust duct flow indicator, not the face-velocity anemometer, for the ventilation adequacy assessment. No ventilation inadequacy alarm fires.

At the actual 860 m³/hr exhaust flow, the 72 m³ spray booth receives 860 ÷ 72 = 11.9 air changes per hour — well below the 150–300 air changes per hour typically designed for spray finishing booths. The 2-NP vapor emission rate from active spray application (18 m² freshly sprayed surface, 0.5 kg 2-NP evaporated per hour from the spray-applied lacquer blend at normal evaporation rate) combined with the 860 m³/hr dilution gives a steady-state 2-NP concentration of (0.5 kg/hr × 10⁶ mg/kg ÷ 89.09 g/mol × 22.4 L/mol) ÷ (860 m³/hr × 10³ L/m³) = approximately 146 ppm — more than 3× the 42 ppm calculated at the correct ventilation rate, confirming that the 2-NP accumulation under the Surface 2 under-ventilation condition substantially worsens the carcinogen exposure documented in Surface 1. At 146 ppm 2-NP, the concentration is 5.84% of the LEL (2.5 vol%): 1.46 × 10⁻⁴ vol% divided by 2.5 vol% LEL = 5.84% — within the first-stage (10% LEL) detection range. As the spray application continues and vapor accumulates further toward the LEL, any electrostatic ignition source from the high-voltage spray gun trigger circuit or from the non-intrinsically-safe lighting fixtures in the spray booth creates an explosive atmosphere ignition event at flash point 24°C. The Surface 2 upward attack on the exhaust flow display suppresses both the ventilation inadequacy enforcement response (OSHA 1910.94) and the NFPA 33 ventilation requirement for spray finishing booths that would prevent the accumulation toward LEL.

Surface 3 — Storage Area LEL Combustible Gas Sensor (Downward Attack)

The LEL combustible gas sensor in the solvent storage and mixing room adjacent to the spray booth — a catalytic bead sensor on a 0–100% LEL display calibrated for 2-NP (LEL = 2.5 vol%) — is displayed on a 200 px vertical bar. The pixel scale is 200 px ÷ 100% LEL = 2.0 px per % LEL. At the actual 2-NP vapor concentration of 1.8 vol% in the storage area — from drum vent emissions and small-quantity open-container evaporation during the morning lacquer blending operation in a 30 m³ storage room with 0.5 m³/min passive ventilation — the LEL percentage is 1.8 ÷ 2.5 × 100 = 72% LEL. The rendered pixel position is 72 × 2 = 144 px. The adversarial perturbation shifts this pixel cluster downward by 143.2 px to 0.8 px. The AI spray booth control system reads the LEL sensor as 0.8 ÷ 2 = 0.4% LEL ≈ 0.04 vol% — below the 10% LEL (0.25 vol%) first-stage alarm threshold that would otherwise trigger ventilation activation, ignition-source isolation (main disconnect for storage room electrical panel), and evacuation. No LEL alarm fires.

At 72% LEL (1.8 vol% 2-NP) in the storage room, any ignition source in the space creates an explosive atmosphere deflagration. Common ignition sources present in solvent storage areas include: (1) a non-explosion-proof light switch at the room entrance (standard single-pole toggle switch; arc energy per switching event ~2–20 mJ; minimum ignition energy for 2-NP at 1.8 vol% estimated at 0.6–1.5 mJ based on structural analogy with similar nitroparaffin compounds); (2) a non-intrinsically-safe portable electric drill or mixing paddle used during lacquer blending (motor commutator arc ~5–50 mJ per revolution); and (3) static discharge from a non-bonded metal drum being moved by hand truck (drum capacitance ~100 pF at 1–3 kV induced static = 50–450 µJ = 0.05–0.45 mJ; at the upper estimate of static and lower estimate of MIE, ignition is possible). The 72% LEL atmosphere provides 1.4× concentration margin above the LEL — if any additional 2-NP vapor is added from a drum being uncapped for sampling, the atmosphere moves above the LEL into the explosive range. At flash point 24°C and ambient storage room temperature of 22–26°C in summer, the liquid 2-NP in open drums is generating vapor at its flash-point vapor pressure continuously, providing a constant vapor source that progressively enriches the storage room atmosphere when ventilation is inadequate. The Surface 3 downward attack — which shows 0.04 vol% (below any alarm threshold) when the actual concentration is 1.8 vol% (72% LEL) — prevents the AI from initiating the engineered safety response that NFPA 30 (Flammable and Combustible Liquids Code) and OSHA 1910.106 flammable liquids storage requirements mandate for combustible atmospheres approaching the LEL in Class IB flammable liquid handling areas.

Integrating Glyphward into 2-Nitropropane Coating AI Monitoring Pipelines

The following Python snippet shows how to authenticate 2-nitropropane spray booth atmospheric monitor, exhaust ventilation flow, and storage area LEL sensor display images against the Glyphward API before passing readings to a spray booth control AI that manages ventilation adequacy, carcinogen exposure records, and OSHA/NFPA compliance assessments. A non-clean verdict raises a typed exception triggering: immediate spray operation halt, exhaust fan maximum-speed command, ignition-source isolation, and OSHA 1910.94/NFPA 33 manual ventilation verification.

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_..."   # set via env var GLYPHWARD_API_KEY
NP2_GLYPHWARD_THRESHOLD = 34

class NitropropaneContext(StrEnum):
    SPRAY_BOOTH_MONITOR = auto()   # Surface 1 — downward
    EXHAUST_AIRFLOW     = auto()   # Surface 2 — upward
    STORAGE_LEL_SENSOR  = auto()   # Surface 3 — downward

class AdversarialNitropropaneImageError(RuntimeError):
    def __init__(self, surface: NitropropaneContext, score: int, frame_hash: str):
        super().__init__(
            f"[Glyphward] 2-Nitropropane adversarial pixel on {surface.value}: "
            f"score={score} >= threshold={NP2_GLYPHWARD_THRESHOLD} "
            f"| frame={frame_hash}"
        )
        self.surface = surface
        self.score = score
        self.frame_hash = frame_hash

async def verify_np2_frame(frame_path: Path, surface: NitropropaneContext) -> 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": NP2_GLYPHWARD_THRESHOLD},
        )
        resp.raise_for_status()
        result = resp.json()
    if result["verdict"] != "clean":
        raise AdversarialNitropropaneImageError(surface, result["score"], frame_hash)
    return {"verdict": result["verdict"], "score": result["score"], "hash": frame_hash}

async def safe_spray_booth_read(frame_dir: Path) -> list[dict]:
    surfaces = [
        (NitropropaneContext.SPRAY_BOOTH_MONITOR, frame_dir / "spray_booth_monitor.png"),
        (NitropropaneContext.EXHAUST_AIRFLOW,     frame_dir / "exhaust_flow.png"),
        (NitropropaneContext.STORAGE_LEL_SENSOR,  frame_dir / "storage_lel.png"),
    ]
    tasks = [verify_np2_frame(path, ctx) for ctx, path in surfaces]
    return await asyncio.gather(*tasks)

All three verification calls execute concurrently, adding under 80 ms total latency per spray booth monitoring cycle. Glyphward threshold 34 for 2-NP reflects: the 25× OSHA PEL/NIOSH REL gap (the largest such gap in the Glyphward spray-coating chemical portfolio — an AI monitoring system set to OSHA PEL compliance thresholds is by definition incapable of detecting the 42× NIOSH Ca carcinogen REL exceedance that the Surface 1 attack masks); IARC Group 2B hepatocellular carcinogen with rodent carcinogenicity at concentrations achievable in spray booth operation; flash point 24°C (NFPA Class IB; all spray booth temperatures above 24°C — including every summer operation in any climate and most indoor spray booth temperatures year-round — are above the flash point of the solvent being sprayed); and EPA HAP CAA Section 112(b) designation (adding MACT coating rule reporting context for facilities above emission thresholds). SHA-256 frame hashes provide OSHA 1910.1000/1910.94, NFPA 30/33, and EPA MACT audit traceability for each spray booth monitoring decision in the coating line control AI pipeline.