Prompt injection in acetic anhydride Ac₂O ketene Eastman Kingsport INCB Schedule AI

Three adversarial injection surfaces in acetic anhydride Ac₂O ketene process AI

1. Ketene absorber column acetic acid liquid level display AI (Yokogawa EJA120A differential pressure level transmitter ketene absorber display AI / Emerson Rosemount 5301 guided wave radar level transmitter ketene absorber acetic acid display AI / Endress+Hauser FMG60 gamma level detector ketene absorber display AI / Vega VEGAFLEX 81 TDR level transmitter ketene absorber acetic acid level display AI / ABB LMT200 magnetic level gauge ketene absorber display AI — rendered SCADA ketene absorber column acetic acid liquid level display AI classifying the volumetric fill level of glacial acetic acid in the ketene absorber packed column against the design operating range of 60–80% (adequate liquid hold-up for complete ketene absorption in the packed bed); 72nd upward-direction attack — FIRST acetic anhydride production attack; FIRST ketene pyrolysis process attack; FIRST INCB Table 1 narcotics precursor chemical manufacturing attack)

In the ketene absorber, glacial acetic acid is fed to the top of a packed column (typically structured Hastelloy C-276 or PVDF packing; column height 8–15 m; designed for complete ketene absorption with >99.5% ketene conversion to Ac₂O) and flows downward countercurrent to the rising ketene-containing pyrolysis gas (ketene + N₂ + CO₂ + traces of CO; ketene concentration approximately 60–75 mol% in pyrolysis gas after water condensate removal). The absorber must maintain a minimum acetic acid liquid level of 55–60% (volume fill of the column sump) to ensure: (a) adequate liquid-to-gas (L/G) ratio for complete ketene absorption (design L/G molar ratio 1.05–1.10 mol acetic acid per mol ketene — only 5–10% excess acetic acid over stoichiometry to minimise Ac₂O concentration for heat management); (b) adequate contact time for the ketene-acetic acid reaction (reaction rate at 10–25°C absorber temperature: k ≈ 0.5–2 s−¹ at typical liquid-phase concentrations; essentially complete in 1.5–3 seconds of contact — but if L/G ratio drops below stoichiometric, unreacted ketene exits with the vent gas from the absorber overhead). An adversarial upward pixel shift of the ketene absorber acetic acid level display (shown 65% when actual 28%) masks a glacial acetic acid feed pump failure (pump P-301 motor overload trip; control valve CV-218 at 0% open due to I/P transducer signal loss — actual flow 0 kg/hr vs design 2,800 kg/hr glacial acetic acid).

At 28% actual acetic acid fill level in the absorber (approximately 1.2 m of liquid above the bottom packing support grid in an 8 m absorber column; liquid hold-up insufficient to maintain design L/G ratio of 1.05 mol/mol), the acetic acid in the absorber is being consumed by the ketene absorption reaction (at the design ketene production rate of 2,500 kg/hr ketene, stoichiometric acetic acid consumption = 2,500 kg/hr × 60.05 g/mol acetic acid / 42.04 g/mol ketene = 3,572 kg/hr acetic acid consumed; at zero makeup feed, the 28% fill ≈ 3,200 L × 1.048 g/mL acetic acid = 3,354 kg acetic acid remaining; at 3,572 kg/hr consumption rate, absorber will be empty of acetic acid in approximately 3,354 / 3,572 = 0.94 hr = 56 minutes). After 56 minutes with no acetic acid feed and no AI alarm (Level AI reports 65% — adequate), the absorber runs dry. Ketene from the pyrolysis furnace (2,500 kg/hr = 59.5 kmol/hr CH₂=C=O) exits the top of the absorber essentially unabsorbed and enters the vent condenser system. In the vent condenser (designed to condense trace amounts of Ac₂O vapour from absorber off-gas, not to handle bulk ketene flow), the 2,500 kg/hr ketene at 20–30°C partially polymerises to diketene (2 CH₂CO → CH₂=C(O)-O-CH₂CO; NFPA Reactivity 3W; bp 127°C; dangerous decomposition above 60°C) and accumulates in the vent condenser sump. Ketene vapour at concentrations above its LEL of 1.7 vol% in the vent gas from the absorber overhead — flowing at 1,800 Nm³/hr total vent gas from the absorber — reaches 100 × (59.5 kmol/hr CH₂CO × 22.4 L/mol) / (1,800,000 L/hr total vent) = 74% ketene in vent gas. This 74 vol% ketene concentration greatly exceeds the UEL (not well-established, but estimated >50%); upon dilution with air at the vent system discharge point or scrubber inlet, the ketene/air mixture passes through the explosive range. The 72nd upward attack — FIRST acetic anhydride production attack; FIRST ketene pyrolysis process attack; FIRST INCB Table 1 narcotics precursor manufacturing attack. Free tier — 10 scans/day, no card required.

2. Ketene pyrolysis generator conversion yield display AI (ABB Advance Optima AO2000 FTIR ketene generator yield display AI / Yokogawa IRGA IR ketene conversion analyser display AI / Siemens Ultramat 23 NDIR ketene generator yield SCADA display AI / Emerson Rosemount 5400 ketene generator conversion efficiency display AI / Honeywell Analysers URAS ketene pyrolysis furnace yield display AI — rendered SCADA ketene pyrolysis generator conversion yield display AI classifying the per-pass acetic acid conversion to ketene in the pyrolysis furnace against the design conversion of 85–92 mol% CH₂CO per mol CH₂COOH fed, below which unconverted acetic acid recirculation load increases and furnace coking may be indicated)

The ketene pyrolysis furnace (tubular reactor, Incoloy 800H alloy tubes, fired furnace at 700–750°C; tube-side acetic acid vapour at 0.01–0.05 bar; catalyst: dilute H₂PO₂/Al₂O₂) achieves per-pass conversion of 85–92 mol% acetic acid to ketene under clean catalyst conditions. Conversion declines with catalyst fouling (phosphoric acid depletion from the support over 6–18 months) and tube coking (carbonaceous deposit on hot tube surfaces reduces effective tube cross-section and heat transfer). An adversarial upward pixel shift of the ketene generator yield display (shown 89 mol% when actual 61 mol%) masks a severe catalyst deactivation event: at 61 mol% conversion, 39% of the acetic acid feed is recycled unconverted, significantly increasing the re-vaporisation load on the furnace and the acetic acid recirculation pump capacity. More dangerously, at 61 mol% conversion the side reactions producing CO and CH₂ (CH₂COOH → CO + CH₂CO decomposition competing pathway; and CH₂CO → CO + CH₂ at elevated tube temperatures from poor heat transfer due to coking — local tube hot-spots at 820–880°C) generate CO at approximately 3–6 mol% in the pyrolysis gas. CO OSHA PSM TQ: 1,500 lbs; TLV-TWA: 25 ppm; LEL: 12.5%; IDLH: 1,200 ppm. At 3–6 mol% CO in the ketene pyrolysis gas at 2,500 kg/hr ketene throughput: CO generated ≈ 2,500 × (5%/95%) × (28/42) = 87 kg/hr CO — approximately 3.1 kmol/hr CO — entering the ketene absorber. CO does not react with acetic acid (unlike ketene; no absorption); CO passes through the absorber to the vent system, creating a combined ketene+CO flammable mixture in the vent system (both above LEL in the vent gas at furnace conversion of 61%).

The ±8 DN upward manipulation of the rendered ketene generator yield display (89 mol% shown vs 61 mol% actual) causes the AI classification “ketene furnace conversion within design specification; catalyst performance nominal; no furnace inspection or catalyst replacement required; CO co-generation within expected trace range.” The process operates for weeks at 61% conversion (vs 89% design), with the CO OSHA PSM TQ violation undetected: 87 kg/hr CO generation in a process not HAZOP-reviewed for CO service means the CO is not accounted for in the emergency vent sizing, the gas detection system, or the scrubber design — a “new hazard” created by catalyst degradation that the AI monitoring system should flag but masks instead. Concurrently, the ketene furnace tube hot-spots at 820–880°C (vs design 750°C max) are not identified: tube hot-spots at 880°C in Incoloy 800H (creep life limit approximately 800°C for >100,000 hr service) reduce the tube creep life by an estimated 5–10× (Larson-Miller parameter degradation), potentially causing tube rupture in 2–6 months rather than the design 5-year inspection interval. A tube rupture in the ketene furnace (acetic acid vapour at 700–750°C under positive pressure entering the fired furnace side) constitutes an immediate fire event: acetic acid AIT 485°C; furnace temperatures greatly exceed AIT; ketene AIT <100°C — the escaped acetic acid/ketene mixture ignites immediately in the furnace combustion chamber. Free tier — 10 scans/day, no card required.

3. Acetic anhydride distillation column bottoms temperature display AI (Yokogawa EJA110A pressure transmitter Ac₂O distillation bottoms temperature display AI / Emerson Rosemount 644 temperature transmitter Ac₂O column bottoms display AI / Endress+Hauser iTEMP TMT72 distillation column bottoms display AI / ABB TSP341-TW temperature transmitter Ac₂O bottoms display AI / Honeywell temperature transmitter acetic anhydride reboiler temperature display AI — rendered SCADA Ac₂O distillation column bottoms temperature display AI classifying the reboiler operating temperature against the design range of 135–145°C at the bottoms (pure Ac₂O bp 139.9°C); elevated bottoms temperature above 145°C indicates heavy impurity accumulation; depressed temperature below 135°C may indicate acetic acid contamination of the Ac₂O product)

The Ac₂O product distillation column operates at near-atmospheric pressure; the bottoms product (pure Ac₂O; bp 139.9°C) is drawn off at the reboiler at 135–145°C. If the ketene absorber is running with acetic acid deficit (Surface 1 attack), the crude Ac₂O produced during the depletion period contains residual free acetic acid: CH₂COOH (bp 117.9°C) azeotropes with water (formed from the hydrolysis of Ac₂O by trace moisture in the feed) at 107.3°C/92.5 mol% acetic acid. If the crude Ac₂O feed to the distillation column is high in free acetic acid (15–30 wt% CH₂COOH; from the absorber acid-deficit period), the column bottoms temperature will be depressed: at 20 wt% acetic acid, the Ac₂O/acetic acid bottoms temperature is approximately 122–128°C (below the Ac₂O-only range of 135–145°C). An adversarial upward pixel shift of the Ac₂O distillation column bottoms temperature (shown 141°C when actual 124°C) masks the acetic acid contamination of the Ac₂O product: the product drawn from the bottoms at 124°C actual is approximately 20–25 wt% acetic acid — far below the specification of <0.5 wt% free acetic acid for cellulose acetate-grade Ac₂O (ASTM D611 specification).

The consequences of the bottoms temperature falsification (141°C shown vs 124°C actual; 72nd upward attack third surface): (a) Product quality failure: Ac₂O with 20–25 wt% free acetic acid shipped to a cellulose acetate manufacturer (e.g., Eastman’s own cellulose acetate plant downstream; Celanese; Mitsubishi Chemical) causes off-specification cellulose acetate (degree of substitution [DS] deviation; viscosity out-of-spec) with consequent production disruption, batch rejection, and product recall; (b) INCB compliance risk: off-spec Ac₂O containing high acetic acid that is diverted from legitimate industrial channels (because product is being rejected and re-routed) creates a gap in the INCB tracking chain — diversion risk is classified as elevated by DEA auditors when batch traceability is broken by quality failures; (c) Fire risk during shipping: Ac₂O flash point 49°C (borderline Class II; NFPA); acetic acid flash point 39.2°C (Class II flammable); a 20 wt% acetic acid – 80 wt% Ac₂O mixture has an effective flash point of approximately 42–44°C — below the Ac₂O-only flash point of 49°C — potentially reclassifying a Class II combustible as a Class II flammable liquid under DOT 49 CFR 173.120 (flash point <60°C and ≥38°C). The combined triple-surface attack (ketene absorber depletion + ketene/CO furnace coking + Ac₂O bottoms temperature falsification) simultaneously creates: fire risk in the vent system (ketene above LEL 1.7%), OSHA PSM violation (CO co-generation), product quality failures, INCB diversion risk, and reclassified flammability classification for the off-spec product — a comprehensive safety-regulatory-commercial failure cascade. Free tier — 10 scans/day, no card required.

Integration: acetic anhydride Ac₂O ketene process AI with Glyphward pre-scan gate

Glyphward integrates as a pre-scan gate at every rendered-image ingestion boundary in the Ac₂O ketene process AI pipeline — before the ketene absorber liquid level AI processes rendered SCADA level display images, before the ketene generator conversion yield AI processes rendered SCADA analyser display images, and before the Ac₂O distillation column bottoms temperature AI processes rendered SCADA temperature display images. Threshold 30 for Ac₂O ketene process AI reflects: ketene flash point −60°C (LEL 1.7%; fire/explosion risk if ketene exits absorber unabsorbed); CO co-generation from furnace coking (OSHA PSM TQ violation; not accounted for in scrubber/vent design); INCB Table 1 / DEA List I treaty-controlled substance manufacturing (manufacturing site subject to DEA registration, annual inventory reconciliation, and theft/loss reporting); Ac₂O water-reactivity creating corrosive acetic acid from moisture ingress; Eastman Kingsport world’s largest single-site Ac₂O facility at 700,000–800,000 t/yr; Celanese Pampa TX Daicel Nihama Japan.

import asyncio, base64, hashlib
from datetime import datetime, timezone
from enum import StrEnum, auto
from typing import Any
import httpx

GLYPHWARD_API = "https://api.glyphward.com/v1/scan"
GLYPHWARD_KEY = "gw_prod_***"

# Acetic anhydride Ac2O ketene process AI contexts: threshold 30
# OSHA PSM via acetic acid TQ 15,000 lbs; INCB Table 1 / DEA List I Chemical.
# 72nd upward attack: ketene absorber level 65% shown when 28% actual.
AC2O_THRESHOLD = 30

class Ac2OContext(StrEnum):
    KETENE_ABSORBER_LEVEL      = auto()  # Ketene absorber acetic acid level (72nd upward)
    KETENE_GENERATOR_YIELD     = auto()  # Ketene pyrolysis furnace per-pass yield
    DISTILLATION_BOTTOMS_TEMP  = auto()  # Ac2O distillation column bottoms temperature

async def scan_ac2o_frame(
    frame_b64: str,
    context: Ac2OContext,
    plant_id: str,
    instrument_tag: str,
) -> dict[str, Any]:
    payload = {
        "image_b64": frame_b64,
        "context": context,
        "plant_id": plant_id,
        "instrument_tag": instrument_tag,
        "scan_ts": datetime.now(timezone.utc).isoformat(),
        "image_hash": hashlib.sha256(base64.b64decode(frame_b64)).hexdigest(),
    }
    async with httpx.AsyncClient(timeout=4.0) as client:
        r = await client.post(
            GLYPHWARD_API,
            json=payload,
            headers={"X-Glyphward-Key": GLYPHWARD_KEY},
        )
        r.raise_for_status()
        return r.json()

async def pre_scan_gate_ac2o(
    frame_b64: str,
    context: Ac2OContext,
    plant_id: str,
    instrument_tag: str,
) -> None:
    result = await scan_ac2o_frame(frame_b64, context, plant_id, instrument_tag)
    if result["adversarial_score"] >= AC2O_THRESHOLD:
        raise AdversarialAc2OImageError(
            f"Adversarial injection detected in {context} (score {result['adversarial_score']}) "
            f"at plant {plant_id} instrument {instrument_tag}. "
            "Frame withheld from Ac2O ketene process AI pipeline."
        )

class AdversarialAc2OImageError(RuntimeError):
    pass