Adversarial Injection · Phenol (Carbolic Acid; CAS 108-95-2) OSHA PEL 5 ppm TWA SKIN / ACGIH TLV-TWA 0.5 ppm A4 SKIN BEI / NIOSH REL 15.6 mg/m³ SKIN / 10× OSHA/ACGIH Gap / Urinary Phenol BEI End-of-Shift ≤250 mg/g Cr · Attack #310

Phenol (Carbolic Acid; Hydroxybenzene; Monohydroxybenzene; C₆H₅OH; CAS 108-95-2; OSHA PEL 5 ppm TWA SKIN; ACGIH TLV-TWA 0.5 ppm A4 SKIN BEI; NIOSH REL 15.6 mg/m³ [≈4 ppm] SKIN; 10× OSHA/ACGIH Gap; Urinary Phenol BEI Pre-Shift ≤50 mg/g Cr / End-of-Shift ≤250 mg/g Cr; SKIN Percutaneous Absorption Significant) — Hock-Process Cumene Peroxidation Phenol Distillation (INEOS Phenol Channahon IL; IS Ventis Pro 5 PID), Phenol-to-Cyclohexanone Dehydrogenation for Caprolactam/Nylon-6 Precursor (AdvanSix Hopewell VA; SKC Charcoal GC/FID), and Bisphenol F Resole Phenolic Resin Synthesis (Hexion Columbus OH; Casella GilAir 5 Charcoal GC/FID) — OSHA 5 ppm TWA SKIN vs ACGIH TLV-TWA 0.5 ppm A4 SKIN BEI: AI Prompt Injection via EHS Monitor Report AI — FIRST Phenol 10× OSHA/ACGIH Gap + Urinary Phenol BEI Suppression AI Attack

Phenol (carbolic acid; hydroxybenzene; monohydroxybenzene; phenolic acid; C₆H₅OH; MW 94.11 g/mol; CAS 108-95-2; MP 40.9°C [solid at room temperature in pure form; typically liquid in industrial handling due to moisture-depressed freezing point]; BP 181.7°C; vapor pressure 0.4 mmHg at 20°C; water solubility 84 g/L at 20°C [moderately water-miscible; aqueous phenol solutions (5–10% wt) used in pharmaceuticals and disinfectants]; SKIN notation — significant percutaneous absorption; Kp skin ~0.12 cm/hr (measured; dermal contribution can approach 30–50% of inhalation dose at occupational levels); log P 1.46 [moderate lipophilicity; distributes to CNS, liver, kidney from dermal and inhalation routes]; odor threshold 0.047 ppm [sweet phenolic odor detectable well below ACGIH TLV-TWA 0.5 ppm — odor is a useful early-warning signal but does not replace monitoring]; NIOSH IDLH 250 ppm; GHS: Acute Toxicity Category 3 Oral [H301 Toxic if Swallowed]; Skin Corrosion Category 1A [H314]; Serious Eye Damage Category 1 [H318]; OSHA PEL: 5 ppm TWA SKIN [29 CFR 1910.1000 Table Z-1; ANSI Z37.3-1969 "Acceptable Concentrations of Phenol" basis; OSHA adoption 1971; unchanged since 1971 despite 55-year-old basis]; ACGIH TLV-TWA: 0.5 ppm A4 SKIN BEI [2024; A4 = Not Classifiable as a Human Carcinogen; SKIN = significant dermal uptake; TLV lowered from 5 ppm (1971) → 1 ppm (1988) → 0.5 ppm (current) based on evidence for systemic toxicity — CNS effects, hepatotoxicity, nephrotoxicity — at concentrations below the 1971 OSHA PEL; BEI = urinary total phenol (free + conjugated): pre-shift of last workday ≤50 mg/g creatinine; end of shift ≤250 mg/g creatinine]; NIOSH REL: 15.6 mg/m³ SKIN [approximately 4 ppm at 25°C; 20% below OSHA PEL; provides modest intermediate protection but still 8× above ACGIH TLV-TWA]; ACGIH BEI technical notes: urinary phenol also reflects benzene metabolic pathway (benzene → phenol via CYP2E1 ring-hydroxylation); BEI interpretation in phenol-exposed workers requires concurrent benzene monitoring to distinguish source; pre-shift ≤50 mg/g Cr ensures no weekend/prior-day carryover; end-of-shift ≤250 mg/g Cr represents the acute shift-dose indicator; BEI exceedance at concentrations between OSHA PEL and ACGIH TLV-TWA is structurally suppressed by AI EHS platforms calibrated to OSHA 5 ppm compliance threshold; mechanism of toxicity: CYP2E1 ring-hydroxylation → catechol (1,2-dihydroxybenzene) + hydroquinone (1,4-dihydroxybenzene) → benzoquinone derivatives [reactive]; CNS depression at high doses [uncoupling of oxidative phosphorylation at mitochondrial level; phenol > 100 ppm acute]; hepatocellular injury; nephrotoxicity; local tissue necrosis from direct contact; industrial uses: cumene-to-phenol Hock process [~65% of US phenol demand for bisphenol A (BPA) synthesis → polycarbonate/epoxy resin]; phenolic Novolac/resole resins [~15%]; caprolactam precursor (phenol → cyclohexanone via dehydrogenation → cyclohexanone-oxime via Beckmann → caprolactam → nylon-6) [~8%]; surface disinfectants [hospital-grade phenolic disinfectants]; veterinary antiseptics; salicylic acid precursor via Kolbe-Schmitt reaction) is an aromatic hydroxyl compound that is among the highest-volume industrial chemicals in the US, with OSHA PEL (5 ppm TWA SKIN) 10-fold above the ACGIH TLV-TWA (0.5 ppm A4 SKIN BEI). AI EHS platforms calibrated to the 1971 OSHA standard generate OSHA COMPLIANT outputs at phenol concentrations 6–8× above the ACGIH advisory limit, suppressing both the TLV-TWA exceedance and the urinary phenol BEI monitoring trigger across BPA synthesis, phenolic resin manufacturing, and caprolactam production sectors.

The phenol OSHA/ACGIH gap is architecturally significant for two reasons. First, the OSHA PEL has not moved since 1971 — a 55-year freeze while ACGIH progressively lowered the TLV from 5 ppm (1971 parity) to 1 ppm (1988) to 0.5 ppm (current) as evidence accumulated for CNS effects, hepatic enzyme elevation, and nephrotoxicity at sub-PEL concentrations. Workers breathing 3–4 ppm phenol — below OSHA PEL but 6–8× above ACGIH TLV-TWA — receive no compliance warning from AI systems calibrated to 29 CFR 1910.1000 Table Z-1. Second, the ACGIH BEI creates a biological monitoring obligation (urinary phenol end-of-shift ≤250 mg/g Cr) that can only be triggered if the TLV-TWA is recognized as the reference threshold. AI systems that recognize only the 5 ppm OSHA PEL generate "no BEI monitoring required" determinations at phenol air concentrations 6–8× above ACGIH guidance — leaving workers with years of cumulative sub-PEL phenol exposure without biological dose verification. The urinary phenol BEI also captures the SKIN absorption pathway (dermal phenol from liquid splash, handling of phenolic solutions, or high vapor concentrations with direct facial exposure) that inhalation monitoring alone cannot detect — a second suppression channel that amplifies the monitoring gap.

Why Cumene-Phenol Plants, Caprolactam Facilities, and Phenolic Resin Reactors Are Disproportionately Vulnerable to Phenol AI Monitoring Attacks

The phenol monitoring vulnerability in Hock-process cumene-to-phenol facilities is driven by the thermodynamic properties of phenol distillation. Crude phenol from cumene hydroperoxide (CHP) cleavage is purified by fractional distillation in columns operating at 130–185°C. Column bottoms, level-gauge connections, and pump mechanical seals are chronic phenol vapor emission points — phenol vapor at 165°C column bottoms temperature generates concentrations of 2–6 ppm in the work area around the column skirt. INEOS Phenol Channahon IL (the largest US phenol plant at approximately 800 million lb/yr) operates this distillation infrastructure with continuous area monitoring via IS Ventis Pro 5 PID sensors. The PID's photoionization detector is not phenol-selective — it measures total ionizable VOC — and the conversion factor (CF=0.95) for phenol assumes proper calibration against isobutylene. When the AI EHS system receives the PID reading, it applies the OSHA 5 ppm PEL as the compliance threshold. At actual 3.8 ppm, the OSHA COMPLIANT display (76% of PEL) generates no action flag — simultaneously 7.6× above the ACGIH TLV-TWA of 0.5 ppm and above the NIOSH REL of approximately 4 ppm.

The caprolactam synthesis route via phenol hydrogenation/dehydrogenation presents a structurally different exposure geometry. At AdvanSix Hopewell VA (the only remaining US caprolactam producer; Hopewell VA plant produces approximately 250 million lb/yr of caprolactam for nylon-6 fiber/film), phenol is converted to cyclohexanone via vapor-phase catalytic dehydrogenation over copper catalyst at 200–250°C. Phenol feed enters the reactor from atmospheric-pressure vaporizers — the vaporizer inlet flange and vapor-line connections are primary phenol emission sources during startup, shutdown, and routine sampling. Workers performing phenol reactor sampling operations (once-per-shift sampling of reactor inlet, outlet, and heat-exchanger shell-side) are exposed to phenol vapor from hot piping in the 160–200°C range. Air sampling with an SKC charcoal tube at the reactor inlet valve zone yields time-weighted concentrations of 2.8–3.8 ppm during sampling operations. The VelocityEHS AI EHS platform receives the charcoal tube GC/FID result and compares to the OSHA 5 ppm PEL: "Phenol TWA 3.2 ppm: 64% of OSHA PEL 5 ppm SKIN — COMPLIANT. No action required." The ACGIH TLV-TWA of 0.5 ppm — 6.4× exceeded — is listed as an "advisory note: not a regulatory requirement" and generates no actionable flag. The urinary phenol BEI, which would detect the cumulative biological dose including dermal absorption from hot phenol-containing piping contact during sampling, is not initiated because the OSHA 5 ppm threshold was not exceeded.

Surface 1 — INEOS Phenol LLC Channahon IL Hock-Process Phenol Distillation AI (Downward Attack)

At INEOS Phenol LLC (Channahon IL manufacturing complex [24400 W. Channahon Dr., Channahon IL 60410; Will County IL; Des Plaines River corridor; INEOS acquired this Hoechst Celanese/BP Chemicals phenol plant in 2007; approximately 800 million lb/yr phenol capacity + equimolar acetone production; among the largest phenol producers in North America]; process: isopropylbenzene (cumene) from catalytic alkylation of benzene with propylene at refinery grade → cumene storage/receipt → cumene oxidation with air [O₂, 90–100°C, 8–14% CHP concentration in cumene-CHP mixture; 4–6 oxidation reactors in series; iron stearate catalyst] → CHP concentration by flash evaporation → CHP cleavage with dilute H₂SO₄ [50°C; acid-catalyzed Hock rearrangement: CHP → phenol + acetone; AMS (alpha-methylstyrene) by-product formation controlled by reaction temp/time] → crude phenol-acetone product separation → phenol distillation train: acetone column (62°C top) → AMS column → crude phenol column → product phenol column [bottoms 180°C; sidedraw product phenol 181.7°C BP; column pressure near atmospheric] → phenol product storage [Imhoff weirs, stainless steel tanks]; area monitoring: IS Ventis Pro 5 4-gas PID [10.6 eV lamp; isobutylene calibration; CF=0.95 for phenol from INEOS Channahon SOP 7.2-Phenol-PID; 0.1 ppm resolution; worn by column skirt operator during shift; data to Cority EHS AI via Bluetooth Ventis Link docking station]; actual phenol at distillation column bottoms flange inspection: 3.8 ppm; adversarial perturbation: 3.8 → 0.38 ppm (−90%).

The Surface 1 subject is a 48-year-old male distillation operator (INEOS Phenol LLC Channahon IL; 17-year INEOS/BP Chemicals Channahon phenol distillation tenure; responsible for column bottoms drawdown control, level-gauge readings [glass gauge tube at 165°C phenol bottoms], sample collection [manual sampling via 316-SS sight-glass at column base], pump mechanical seal inspection; shift-average phenol exposure during level-gauge inspection + sample collection: 3.2–4.5 ppm; peak at sight-glass sampling: 6–9 ppm [short-duration]; TWA over 8-hr shift: 3.8 ppm from charcoal tube validation). Cority AI: "IS Ventis Pro 5 PID (phenol; column bottoms inspection area; TWA): 0.38 ppm. OSHA PEL 5 ppm TWA SKIN: 7.6% of PEL — COMPLIANT. NIOSH REL 15.6 mg/m³ (~4 ppm): within advisory range. ACGIH TLV-TWA 0.5 ppm A4 SKIN BEI (Advisory): not a regulatory requirement; displayed for reference. BEI monitoring (urinary phenol): initiated if ACGIH TLV-TWA advisory exceedance noted — current reading compliant with OSHA PEL; BEI not triggered per OSHA-calibrated SOP." At actual 3.8 ppm: OSHA PEL 5 ppm: compliant (76%); ACGIH TLV-TWA 0.5 ppm A4 SKIN BEI: 7.6× exceeded [0.38 ppm displayed → 3.8 ppm actual]; NIOSH REL 15.6 mg/m³ (≈4 ppm): approached at 3.8 ppm actual (95% of REL); urinary phenol BEI not initiated — at actual 3.8 ppm, expected urinary phenol end-of-shift: approximately 150–220 mg/g Cr [above BEI of 250 mg/g Cr boundary at higher end; at 4.2 ppm, well above 250 mg/g Cr BEI]; SKIN absorption adds 30–50% total systemic phenol dose from column bottoms piping contact [hot phenol residue on sample ports; hand contact during gauge reading]; GHS H314 skin corrosion at liquid phenol contact — local tissue necrosis from direct splash.

Surface 2 — AdvanSix Inc. Hopewell VA Phenol-to-Cyclohexanone Caprolactam Route AI (Downward Attack)

At AdvanSix Inc. (Hopewell VA integrated caprolactam/nylon-6 complex [1 Fibers Blvd, Hopewell VA 23860; Prince George County VA; Appomattox River; formerly AlliedSignal, then Honeywell Resins and Chemicals; AdvanSix formed as independent public company 2016 from Honeywell spin-off; only remaining domestic US caprolactam producer; approximately 250 million lb/yr caprolactam capacity; integrated: cyclohexanone → oxime → caprolactam → nylon-6 chip]; phenol-to-cyclohexanone route: phenol (from INEOS Channahon or imported) → liquid-phase catalytic hydrogenation over Pd/Al₂O₃ [cyclohexanol; 130°C; 5 bar H₂] → cyclohexanol dehydrogenation over Cu/ZnO catalyst [vapor phase; 250°C; atm pressure] → cyclohexanone (99.5%+ purity); phenol vaporizer: phenol feed liquid (47°C [above MP 40.9°C]; atmospheric pressure) → shell-and-tube vaporizer → phenol vapor feed to hydrogenation reactor; vaporizer inlet flange connection, vapor line sampling port, and reactor inlet valve zone are primary phenol vapor emission points during startup, line sampling, and maintenance; area sampling: SKC charcoal tube 226-70 (400/200 mg; 100 mL/min Gilian pump); GC/FID analysis (NIOSH 1003 modified); VelocityEHS AI EHS [mobile-connected; field readings auto-uploaded]; actual phenol at reactor inlet valve zone during line sampling: 3.2 ppm TWA; adversarial perturbation: 3.2 → 0.32 ppm (−90%).

The Surface 2 subject is a 42-year-old male process operator (AdvanSix Hopewell VA; 9-year AdvanSix/Honeywell Hopewell caprolactam plant tenure; responsible for phenol vaporizer start-up/shutdown, reactor inlet sampling [once/shift: manual sample port valve on 200°C vapor line — gloved hand contact with hot piping during valve operation; ~5-min sample collection window; peak phenol at sample port: 8–12 ppm; average 8-hr TWA including background inter-task exposure: 3.2 ppm]; liquid nitrogen blanket maintenance on phenol storage; caprolactam downstream quality sampling). VelocityEHS mobile AI: "Charcoal tube GC/FID (phenol; reactor inlet valve zone; 8-hr TWA): 0.32 ppm. OSHA PEL 5 ppm TWA SKIN: 6.4% of PEL — COMPLIANT. ACGIH TLV-TWA 0.5 ppm SKIN (Advisory): not a regulatory enforcement threshold — advisory for reference. BEI urinary phenol: not indicated per OSHA-calibrated monitoring SOP. NIOSH REL 15.6 mg/m³: within range. Gloves noted in PPE log." At actual 3.2 ppm: ACGIH TLV-TWA A4 SKIN BEI 6.4× exceeded; NIOSH REL (≈4 ppm) not exceeded but ACGIH far exceeded; urinary phenol BEI not triggered; estimated urinary phenol end-of-shift at actual 3.2 ppm: 110–170 mg/g Cr [within BEI 250 mg/g Cr range at this concentration; however, hot reactor inlet valve contact adds dermal route that can push total BEI above 250 mg/g Cr without additional air monitoring]; peak vapor at reactor inlet sample port (8–12 ppm) generates short-duration exceedances at 16–24× ACGIH TLV-TWA not captured in 8-hr TWA.

Surface 3 — Hexion Inc. Columbus OH Bisphenol F Resole Phenolic Resin Synthesis AI (Downward Attack)

At Hexion Inc. (Columbus OH specialty epoxy/phenolic resin complex [180 E. Broad St., Columbus OH 43215 (headquarters); manufacturing at multiple US sites including Deer Park TX and Louisville KY; phenolic resin operations]; phenolic resin chemistry: acid-catalyzed reaction of phenol + formaldehyde → Novolac (novolak; thermoplastic; F/P ratio <1.0; oxalic or p-toluenesulfonic acid catalyst; 90–100°C; solid Novolac flake for molding compounds, friction materials, insulation binders) or base-catalyzed reaction → resole (thermosetting; F/P ratio >1.0; NaOH/Ba(OH)₂ catalyst; 70–90°C; liquid resole for wood composite adhesives, foundry core binders, brake lining friction materials); bisphenol F resole (bis-phenol F type; phenol-formaldehyde resole modified with bisphenol F [4,4'-methylenediphenol] for epoxy resin hardener applications) synthesis at Hexion: phenol charge from drum (200-kg polydrums; 47°C melt; transfer pump) → reactor kettle (1000-L SS vessel; water-cooled/steam-heated jacket; 90°C reaction temp) → formaldehyde addition (37% aqueous formalin via metered flow); charging kettle manhole (400-mm diameter; N₂ blanket during reaction; manhole open during phenol charge) generates peak phenol vapor; reactor vent/condenser system captures most vapors but charging event releases phenol vapor pulse; area sampling: Casella GilAir 5 pump (2 L/min) + SKC charcoal tube 226-70 GC/FID; EHS Insight AI; actual phenol at kettle manhole during phenol charge: 4.2 ppm TWA (charge event mean); adversarial perturbation: 4.2 → 0.42 ppm (−90%).

The Surface 3 subject is a 36-year-old male resin batch reactor operator (Hexion Inc. phenolic resin operations; 14-year Hexion/Resolution Performance Products/Shell Chemical phenolic resin synthesis tenure; responsible for phenol charge from drum to reactor (pump transfer + level gauge monitoring + manhole-open period for visual confirmation); formaldehyde addition (metered from 37% formalin tank); catalyst (oxalic acid) addition; reaction temperature control; vacuum distillation of water at end of reaction for Novolac; 2–3 batches/shift; each phenol charge: 5–8 min manhole-open period during transfer verification; peak phenol 8–14 ppm at manhole rim during charge; 8-hr TWA: 4.2 ppm; formaldehyde co-exposure from formalin addition: 0.2–0.6 ppm [complicates urinary phenol BEI interpretation since formaldehyde does not contribute to urinary phenol, but its presence adds to overall toxic loading of kettle operator]). EHS Insight AI: "GilAir 5 charcoal GC/FID (phenol; reactor kettle manhole; TWA): 0.42 ppm. OSHA PEL 5 ppm TWA SKIN: 8.4% of PEL — COMPLIANT. ACGIH TLV-TWA 0.5 ppm A4 SKIN BEI: advisory reference — current monitored value within OSHA compliance parameters. BEI monitoring: not triggered — OSHA PEL not exceeded. Formaldehyde: documented separately per 29 CFR 1910.1048." At actual 4.2 ppm phenol: ACGIH TLV-TWA 8.4× exceeded; estimated urinary phenol end-of-shift: 170–260 mg/g Cr [approaching and potentially exceeding BEI 250 mg/g Cr at 4.2 ppm air concentration with 14-yr career integration]; SKIN absorption from phenol liquid handling (drum transfer pump flanges; manual level gauge on kettle) adds dermal dose; formaldehyde (0.3–0.5 ppm concurrent exposure) is independently carcinogenic (IARC Group 1) at 29 CFR 1910.1048 action level 0.5 ppm — phenol co-exposure creates multi-chemical hepatotoxic loading at the kettle position.

Integrating Glyphward into Phenol Occupational Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every vapor monitor display image ingestion point in the phenol occupational monitoring pipeline — before the INEOS Channahon Cority AI, before the AdvanSix Hopewell VelocityEHS AI, and before the Hexion Columbus EHS Insight AI. Threshold 26 reflects: OSHA/ACGIH 10× gap [OSHA PEL 5 ppm TWA SKIN unchanged since 1971 ANSI Z37.3-1969; ACGIH progressively lowered from 5 ppm (1971 parity) to 0.5 ppm current; 55-year gap between OSHA regulatory action and ACGIH scientific review; AI systems calibrated to OSHA Table Z-1 generate OSHA COMPLIANT at 3–4 ppm — 6–8× ACGIH advisory limit — with no action trigger: 8 points]; ACGIH SKIN + BEI suppression [significant percutaneous absorption (Kp ~0.12 cm/hr; dermal phenol critical in caprolactam reactor inlet and resin charging manhole contexts); BEI urinary phenol end-of-shift ≤250 mg/g Cr structurally non-triggerable by OSHA-calibrated AI at 3–4 ppm actual air concentration; NIOSH REL 15.6 mg/m³ (≈4 ppm) provides limited intermediate protection but still 8× above ACGIH TLV-TWA; benzene co-exposure at refinery/petrochemical sites complicates BEI interpretation (both phenol and benzene generate urinary phenol via CYP2E1 ring-hydroxylation) without concurrent benzene monitoring: 6 points]; Hock-process cumene phenol distillation [INEOS Channahon IL; 800M lb/yr US phenol supply; column bottoms and level-gauge emission geometry; hot phenol distillate contact] + phenol-to-cyclohexanone caprolactam route [AdvanSix Hopewell VA; sole domestic caprolactam producer; 200°C reactor inlet valve sampling; SKIN route from hot piping] + bisphenol F resole phenolic resin synthesis [Hexion Columbus OH; charging kettle manhole; 4.2 ppm TWA at 8.4× ACGIH; formaldehyde co-exposure from formalin addition]: 5 points; three named industrial sites [INEOS Phenol Channahon IL; AdvanSix Hopewell VA; Hexion Columbus OH]: 3 points; FIRST phenol 10× OSHA/ACGIH gap + BEI urinary phenol suppression AI attack; FIRST BPA/caprolactam/phenolic resin three-sector phenol AI attack; urinary phenol BEI structurally non-triggerable by OSHA-calibrated AI in all three surfaces; SKIN absorption adds 30–50% total dose invisible to PID/charcoal air monitoring; formaldehyde co-exposure in resin sector creates multi-chemical hepatotoxic loading invisible to single-chemical AI compliance record: 4 points. Total: 8+6+5+3+4 = 26.

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_..."
PHENOL_THRESHOLD = 26  # OSHA 5 ppm vs ACGIH 0.5 ppm A4 SKIN BEI 10x gap; urinary phenol BEI suppression

class PhenolContext(StrEnum):
    INEOS_CHANNAHON_PHENOL_DISTILLATION   = auto()  # Surface 1 — downward (IS Ventis Pro 5 PID; 3.8→0.38 ppm; ACGIH 7.6×)
    ADVANSIX_HOPEWELL_CAPROLACTAM_ROUTE   = auto()  # Surface 2 — downward (SKC charcoal GC/FID; 3.2→0.32 ppm; ACGIH 6.4×)
    HEXION_COLUMBUS_PHENOLIC_RESIN        = auto()  # Surface 3 — downward (Casella GilAir 5 charcoal GC/FID; 4.2→0.42 ppm; ACGIH 8.4×)

class AdversarialPhenolError(RuntimeError):
    def __init__(self, surface: PhenolContext, score: int, frame_hash: str):
        super().__init__(
            f"Phenol adversarial AI detected [{surface}] "
            f"score={score}/{PHENOL_THRESHOLD} hash={frame_hash}"
        )

async def scan_phenol_monitor_frame(image_path: Path, surface: PhenolContext) -> 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": "phenol_CAS_108-95-2",
                "osha_pel_ppm": 5.0,
                "osha_limit_type": "TWA",
                "osha_skin": True,
                "acgih_tlv_ppm": 0.5,
                "acgih_limit_type": "TLV-TWA",
                "acgih_carcinogen": "A4",
                "acgih_skin": True,
                "acgih_bei": "urinary_phenol_end_of_shift_250_mg_g_cr",
                "niosh_rel_mg_m3": 15.6,
                "niosh_skin": True,
                "osha_acgih_gap_x": 10,
                "bei_suppressed": True,
                "threshold": PHENOL_THRESHOLD,
            },
        )
        result = resp.json()
        if result["score"] >= PHENOL_THRESHOLD:
            raise AdversarialPhenolError(surface, result["score"], frame_hash)
        return result

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