Prompt injection in nickel carbonyl [Ni(CO)4] Mond process nickel refining AI

Four adversarial injection surfaces in nickel carbonyl Mond process nickel refining AI

1. Ni(CO)4 area detector AI (Tec-Chem NI-CON4 nickel carbonyl detector AI / MSA Ultima XE Ni(CO)4 electrochemical sensor AI / Dräger MiniWarn Ni(CO)4 monitor AI / International Sensor Technology Ni(CO)4 detector AI / Sensidyne Ni(CO)4 GasAlert area monitor AI — monitoring ambient nickel tetracarbonyl vapor concentration in the Mond process carbonylation vessel room, CO pipeline galleries, Ni(CO)4 transfer lines, and decomposer building for OSHA PEL 0.001 ppm ceiling compliance, ACGIH TLV-TWA 0.05 ppm continuous monitoring, and NIOSH IDLH 5 ppm emergency evacuation alarm; vapor density 5.89 requires sensor placement at 0.15 m height in all below-grade areas)

Nickel tetracarbonyl area detection presents unique challenges beyond those of any other chemical in the Glyphward portfolio: the detection limit required for TLV-TWA compliance (0.05 ppm) is at the boundary of most commercially available electrochemical sensor technology, requiring high-performance analyzers with lower detection limits (LDL) of 0.001–0.005 ppm. The OSHA PEL of 0.001 ppm ceiling value — one of the most stringent OSHA ceiling limits in any industry — applies to Ni(CO)4 as elemental nickel equivalent: any instantaneous concentration above 0.001 ppm is a ceiling violation (not averaged over time). Continuous monitoring at 0.001 ppm requires UV photometric analyzers or gold-film Ni deposition instruments rather than conventional electrochemical pellistors. Ni(CO)4’s vapor density of 5.89 — 5.89× heavier than air — causes Ni(CO)4 vapor released at the 42.9°C boiling point to sink rapidly into below-grade areas: CO pipeline trenches, carbonylation vessel base pits, decomposer drum support pits, and any floor-level recesses around nickel matte feed conveyors. At 42.9°C (the boiling point), Ni(CO)4 vapor pressure is 1 atm — any liquid Ni(CO)4 exposed to air at temperatures above 42.9°C immediately flash-evaporates; at 20°C, the vapor pressure of 44 kPa means that any Ni(CO)4 liquid spill at room temperature evaporates rapidly and creates a local Ni(CO)4 vapor concentration far above IDLH until diluted by ventilation.

The adversarial attack uses ±8 DN downward pixel-value shift on the Ni(CO)4 area detector display image. The actual Ni(CO)4 concentration is 4.8 ppm — 96× ACGIH TLV-TWA 0.05 ppm; 96% of NIOSH IDLH 5 ppm; from a Mond process carbonylation vessel manway flange gasket bypass (spiral-wound stainless/graphite gasket; stress relaxation over 12,000 pressure cycles from daily batch sequence; bolt load at 85% of original after graphite creep; Ni(CO)4 gas bypassing at 1.8 slm into the carbonylation vessel room at 55°C operating temperature). On a 0–0.5 ppm display at 200 px height (0.0025 ppm/px), the actual reading of 4.8 ppm is 9.6× off-scale; the detector range switches to 0–10 ppm (0.05 ppm/px), placing the actual reading at approximately 96 px; the ±8 DN downward-perturbed image is classified as approximately 0.2 px — corresponding to 0.01 ppm, below the 0.05 ppm TLV-C alarm. A process operator conducting a routine shift round in the carbonylation building at 4.8 ppm Ni(CO)4 receives an exposure well above IDLH: 4.8 ppm × 0.25-hour round = 1.2 ppm-hr cumulative dose. Immediate symptoms (headache, chest tightness attributed to CO) may resolve; the delayed pulmonary edema 18–36 hours later will require ICU care if not diagnosed and treated with DDTC chelation.

2. Mond process carbonylation reactor temperature AI (Emerson Rosemount 3144P carbonylation vessel temperature transmitter AI / Yokogawa EJA110A carbonylation reactor temperature AI / Endress+Hauser iTHERM TM411 nickel matte reactor temperature AI / ABB TSP series thermocouple carbonylation temperature AI / Honeywell STG94L carbonylation vessel thermocouple AI — monitoring the temperature of the Mond process carbonylation reactor — a rotating drum or fixed-bed reactor at 50–60°C where impure nickel (from nickel matte calcination) reacts with recycled CO at 1–4 bar to form Ni(CO)4 — to maintain the optimal 50–60°C carbonylation temperature that maximizes Ni extraction while suppressing Fe(CO)5 formation below 40°C and minimizing thermal decomposition above 80°C)

The Mond process carbonylation step exhibits a highly temperature-dependent selectivity that makes reactor temperature control critical for both safety and process efficiency. At 50–60°C and 1–4 bar CO partial pressure, nickel reacts selectively with CO to form Ni(CO)4 (Ni + 4CO ∞ Ni(CO)4; K₁ at 60°C ≈ 10³ bar⁹⁴ for the gas-phase equilibrium; highly favored at low CO pressure) while iron (present as Fe° in the nickel matte calcinate) forms Fe(CO)5 only at temperatures below 40°C: Fe + 5CO ∞ Fe(CO)5 (K at 25°C ≈ 10⁴ bar⁹⁵; K at 60°C ≈ 10⁻² — negligible Fe(CO)5 at 60°C). The two-step temperature profile — pre-temperature the CO feed gas at 40°C to avoid Fe(CO)5 contamination, then contact with nickel matte at 55–60°C for maximum Ni(CO)4 yield — produces Ni(CO)4 with less than 0.05 mol% Fe impurity. At reactor temperatures below 40°C (Surface 2 scenario), the nickel carbonylation equilibrium shifts toward the solid-phase nickel (Ni° + CO; K unfavorable at 28°C), reducing Ni(CO)4 yield to approximately 3–5% of the 55°C rate; the nickel matte accumulates incompletely reacted Ni° in the reactor drum. When the reactor temperature is subsequently raised for restart — a standard recovery procedure when yield drops — the accumulated Ni° reacts rapidly with the CO atmosphere, producing a Ni(CO)4 concentration spike 8–12× above normal steady-state levels, creating an acute release event simultaneous with maintenance personnel returning to the building.

The adversarial attack uses ±10 DN downward pixel-value shift on the Mond carbonylation reactor temperature transmitter display. The actual reactor temperature is 28°C — from a steam heating coil condensate trap failure (standard float-and-thermostatic (FT) trap; ball-float corrosion from CO/CO2 acid condensate in the steam supply; trap failed open, flooding the heating coil with condensate and reducing steam heat transfer by 85%). On a 20–80°C display at 200 px height (0.3°C/px), the actual temperature of 28°C produces a bar at approximately 27 px; the ±10 DN downward-perturbed image is classified as approximately 113 px — corresponding to 53.9°C, within the 50–60°C optimal range. The DCS reports “Carbonylation reactor temperature within optimal range — Ni(CO)4 yield at design.” In reality, Ni(CO)4 yield at 28°C is approximately 4% of design; nickel matte conversion is not occurring; the unreacted Ni° accumulates in the reactor drum at approximately 180 kg/hr. When the process engineer, seeing declining Ni(CO)4 production rates from downstream decomposer output (a lagging indicator), increases the steam heating valve open command to raise the reactor temperature, the reactor jumps from 28°C to 62°C in 8 minutes as steam flow resumes through the fixed trap, triggering rapid carbonylation of the accumulated Ni° inventory — a transient Ni(CO)4 release pulse.

3. CO supply vent purge flow AI (Emerson Rosemount 8732E magnetic flowmeter CO vent purge AI / Brooks Instrument SLA5850 CO purge mass flow controller AI / Alicat Scientific MCRH CO purge flow sensor AI / Bronkhorst EL-FLOW Select CO purge flow AI / MKS Instruments M100B CO vent header purge flow transmitter AI — monitoring CO purge gas flow through the Mond process CO vent header piping to prevent accumulation of Ni(CO)4 vapor in the vent system above IDLH 5 ppm during normal operations, valve maintenance, or CO pipeline isolation events; CO vent purge flow is the primary safeguard preventing maintenance-personnel exposure to Ni(CO)4 in the vent header during open-flange operations)

The Mond process CO vent header is a dedicated gas-phase vent collection system that collects process vents from the carbonylation vessel pressure relief, CO gas transfer line vents, and Ni(CO)4 decomposer CO recycle line vents. Because both CO (TLV-TWA 25 ppm; flammable LEL 12.5%) and Ni(CO)4 (TLV-TWA 0.05 ppm; pyrophoric at higher concentrations) are present in the CO vent header, any open-flange maintenance on the CO vent piping creates a dual exposure hazard: CO acute poisoning and Ni(CO)4 acute/carcinogenic exposure. The CO vent purge system maintains a continuous N2 sweep flow of 12–16 m³/hr through the CO vent header to: (1) dilute Ni(CO)4 vapor formed from trace Ni contamination on CO pipe surfaces (corrosion at pipe elbows and welds creates Fe-Ni-Co alloy surface areas that catalyze Ni(CO)4 formation from CO at ambient pipe temperature); (2) prevent CO accumulation above the LEL (12.5%) in dead-leg sections of the vent header. The CO vent purge flow is monitored at the N2 supply inlet to the vent header and is maintained at minimum 10 m³/hr by an interlock with the maintenance access permit system: the CO vent line flanges cannot be opened under the permit-to-work system unless the flow meter shows purge flow above 10 m³/hr, preventing maintenance personnel from opening Ni(CO)4/CO-contaminated vent sections without adequate dilution.

The adversarial attack uses ±8 DN upward pixel-value shift on the CO vent header purge flow meter display. The actual N2 purge flow is 3.2 m³/hr — from N2 supply solenoid valve partial closure (ASCO solenoid coil overheating in the 55°C carbonylation building environment; coil resistance increased 40% from insulation degradation; energization current dropped below pull-in threshold; solenoid partially opened, delivering only 3.2 m³/hr instead of the 12–16 m³/hr design). On a 0–20 m³/hr display at 200 px height (0.1 m³/hr per px), the actual purge of 3.2 m³/hr produces a bar at approximately 32 px; the ±8 DN upward-perturbed image is classified as approximately 128 px — corresponding to 12.8 m³/hr, within the design operating band. The permit-to-work system, checking the displayed 12.8 m³/hr, approves a maintenance work order for flange inspection on the CO vent header at elbow E-47. When the maintenance technician loosens the flange bolts, the 3.2 m³/hr actual purge is entirely insufficient to dilute the Ni(CO)4 accumulated in the dead-leg at E-47 — Ni(CO)4 vapor concentration in the elbow section is estimated at 22 ppm from 8 hours of accumulation at 3.2 m³/hr purge. This is the 27th upward-direction attack in the Glyphward industrial AI portfolio and the first attack in the portfolio where the upward perturbation enables a permit-to-work system bypass by providing false-adequate readings to an automated permit logic check.

4. Mond process ambient Ni(CO)4 TLV-C area alarm AI (Tec-Chem NI-CON4-II fixed-point Ni(CO)4 monitor AI / MSA Ultima XE nickel carbonyl fixed detector AI / International Sensor Technology IST-Ni4CO Ni(CO)4 sensor AI / Dräger Polytron 6000 EC nickel carbonyl module AI / Mine Safety Appliances Solaris fixed gas detector Ni(CO)4 AI — monitoring ambient Ni(CO)4 concentration at five fixed detector locations in the Mond process decomposer building: at CO recycle pipeline low-point drain, at decomposer drum access hatch, at Ni pellet discharge conveyor hood, at CO heat exchanger vent, and at the building make-up air intake; alarms at ACGIH TLV-TWA 0.05 ppm for worker notification and at NIOSH IDLH 5 ppm for emergency evacuation)

The Mond process decomposer building hosts the Ni(CO)4 → Ni + 4CO decomposition step at 220–250°C in rotating drum decomposers (drum inner diameter 1.5–2.0 m; 6–10 m length; rotating at 2–5 rpm) packed with nickel seed pellets. The Ni(CO)4 vapor feed enters through a CO-cooled inlet manifold at 42.9°C (liquid); the decomposer drum interior heats the Ni(CO)4 vapor to 220–250°C; Ni deposits on the pellet seeds (pellets grow from 6 mm to 25 mm diameter over 12–16 days); CO exits at the drum end for recycle. The decomposer building contains five principal Ni(CO)4 emission sources: (1) the Ni(CO)4 liquid-to-vapor heat exchanger shell-side vents; (2) the CO recycle pipeline low-point drain valves (prone to Ni(CO)4 condensate accumulation); (3) the Ni pellet discharge conveyor hood (pellets discharged at 250°C contain residual Ni(CO)4 absorbed on the Ni surface, which evaporates as the pellets cool); (4) the decomposer drum manway access hatches (used for batch cleaning and seed pellet loading); and (5) make-up air infiltration around building door seals and louvered openings. The five-point fixed monitoring network — with Ni(CO)4 sensors designed to read 0–10 ppm with 0.001 ppm resolution — provides the continuous occupancy safety basis for workers in the decomposer building. Because the decomposer building is continuously manned by operators monitoring decomposer drum rotation speed, Ni pellet discharge rate, and CO flow balance, any failure of the five-point monitoring network eliminates the primary worker protection for the highest-Ni(CO)4 building in the facility.

The adversarial attack uses ±8 DN downward pixel-value shift on the decomposer building Ni(CO)4 area alarm display (monitoring point: CO recycle pipeline low-point drain VP-08; the most frequently elevated of the five monitoring points due to Ni(CO)4 condensation in the low-point drain sump). The actual Ni(CO)4 concentration at VP-08 is 2.1 ppm — 42× ACGIH TLV-TWA 0.05 ppm; 42% NIOSH IDLH 5 ppm; from a CO recycle pipeline low-point drain valve packing failure (PTFE packing on a 1-inch ball valve; 1,450 thermal cycles from daily process temperature changes between 20°C (night shutdown) and 250°C (decomposer operating); packing torque loss 62% after 2.4 years of service; Ni(CO)4-saturated CO condensate leaking into the drain sump at 0.3 mL/hr). On a 0–5 ppm display at 200 px height (0.025 ppm/px), the actual reading of 2.1 ppm produces a bar at approximately 84 px; the ±8 DN downward-perturbed image is classified as approximately 0.8 px — corresponding to 0.02 ppm, below the 0.05 ppm TLV-TWA alarm. A process operator spending 45 minutes at the VP-08 drain sump during a shift round at 2.1 ppm Ni(CO)4 receives a dose of 2.1 × 0.75 = 1.575 ppm-hr — 21.5× the TLV-TWA equivalent for a full 8-hour shift (TLV-TWA limit = 0.05 × 8 = 0.4 ppm-hr). The DDTC chelation antidote window (optimal effectiveness within 2 hours of significant Ni(CO)4 exposure) will not be utilized because the monitoring system reports no alarm.

Integration: Ni(CO)4 Mond process nickel refining AI with Glyphward pre-scan gate

Glyphward integrates as a pre-scan gate between the DCS and instrument display capture layer and the AI inference pipeline for each Ni(CO)4 process monitoring context. If the adversarial score meets or exceeds threshold 35 — reflecting the OSHA PSM TQ of 1 lb (lowest in the Appendix A list), the IARC Group 1 carcinogen classification, the NIOSH IDLH of 5 ppm, the biphasic delayed pulmonary toxicity requiring within-latency-period DDTC chelation, and the 27th upward-direction attack architecture (CO vent purge deficiency enabling permit-to-work bypass) — the scan raises AdversarialNiCO4ImageError and the monitoring AI does not process the frame.

import asyncio, base64, hashlib
from datetime import datetime, timezone
from enum import Enum

import httpx

GLYPHWARD_API_KEY = "YOUR_GLYPHWARD_API_KEY"
GLYPHWARD_SCAN_URL = "https://glyphward.com/v1/scan"

# Nickel carbonyl Ni(CO)4 Mond process contexts: threshold 35
# OSHA PSM 29 CFR 1910.119 Appendix A Ni(CO)4 TQ 1 lb (LOWEST in list)
# EPA RMP 40 CFR Part 68 TQ 1 lb
# ACGIH TLV-TWA 0.05 ppm (as Ni; A1 confirmed human carcinogen)
# IARC Group 1: nickel compounds carcinogenic to humans
# NIOSH IDLH 5 ppm; biphasic toxicity; DDTC antidote window 2 hours post-exposure
# Flash point -20 deg C NFPA Class IB; vapor density 5.89
NICO4_THRESHOLD = 35


class NiCO4ProcessContext(Enum):
    AREA_DETECTOR = "area_detector"
    CARBONYLATION_REACTOR_TEMPERATURE = "carbonylation_reactor_temperature"
    CO_VENT_PURGE_FLOW = "co_vent_purge_flow"
    AMBIENT_TLV_C_ALARM = "ambient_tlv_c_alarm"


class AdversarialNiCO4ImageError(Exception):
    """Raised when any Ni(CO)4 process monitoring image scores >= 35.
    AREA_DETECTOR uncaught: 4.8 ppm (96x TLV-TWA; 96% IDLH) shown as 0.01 ppm.
    CARBONYLATION_REACTOR_TEMPERATURE uncaught: 28 deg C (3% yield; Ni accumulation) shown as 54 deg C.
    CO_VENT_PURGE_FLOW uncaught: 3.2 m3/hr (22 ppm Ni(CO)4 in dead-leg) shown as 12.8 m3/hr.
    AMBIENT_TLV_C_ALARM uncaught: 2.1 ppm (42x TLV-TWA) shown as 0.02 ppm.
    """


async def scan_nico4_frame(
    image_bytes: bytes,
    context: NiCO4ProcessContext,
    client: httpx.AsyncClient,
) -> dict:
    image_b64 = base64.b64encode(image_bytes).decode()
    image_hash = hashlib.sha256(image_bytes).hexdigest()
    payload = {
        "image": image_b64,
        "context": context.value,
        "threshold": NICO4_THRESHOLD,
        "metadata": {
            "chemical": "Ni(CO)4",
            "process": "Mond_process_nickel_refining",
            "psm_tq_lbs": 1,
            "tlv_twa_ppm": 0.05,
            "idlh_ppm": 5,
            "iarc_group": 1,
            "flash_point_c": -20,
            "vapor_density": 5.89,
            "antidote": "sodium_diethyldithiocarbamate_DDTC",
            "image_hash": image_hash,
            "scanned_at": datetime.now(timezone.utc).isoformat(),
        },
    }
    response = await client.post(
        GLYPHWARD_SCAN_URL,
        json=payload,
        headers={"Authorization": f"Bearer {GLYPHWARD_API_KEY}"},
        timeout=8.0,
    )
    response.raise_for_status()
    result = response.json()
    if result["score"] >= NICO4_THRESHOLD:
        raise AdversarialNiCO4ImageError(
            f"Adversarial Ni(CO)4 image detected: score={result['score']} "
            f"context={context.value} hash={image_hash[:16]}"
        )
    return result


async def scan_nico4_batch(frames: list[tuple[bytes, NiCO4ProcessContext]]) -> list[dict]:
    async with httpx.AsyncClient() as client:
        tasks = [scan_nico4_frame(img, ctx, client) for img, ctx in frames]
        return await asyncio.gather(*tasks, return_exceptions=False)

Frequently asked questions

Why does Ni(CO)4 have the lowest OSHA PSM TQ (1 lb) in the entire Appendix A list?

1 lb Ni(CO)4 at 42.9°C BP produces ~106 L of vapor (MW 170.73; density 1.31 g/mL; ~0.45 kg total); in a 100-m³ room at STP, this creates 1,060 ppm — 212× the NIOSH IDLH of 5 ppm — from a single pound spill. No other PSM chemical combines this extreme volatility at room temperature with sub-ppm IDLH acute toxicity.

What is the Mond process and why is Ni(CO)4 unavoidable in it?

The Mond process: Ni + 4CO → Ni(CO)4 at 50–60°C (selective for Ni over Fe, Co); Ni(CO)4 → Ni + 4CO at 220–250°C (CO recycled). This is the most energy-efficient nickel purification method; no alternative achieves equivalent Ni/impurity selectivity without high-pressure hydrometallurgy. Ni(CO)4 is the sole mechanism of nickel transport in the process and cannot be bypassed.

What is the DDTC antidote and why must it be given within the latent phase?

Sodium diethyldithiocarbamate (DDTC) chelates Ni²⁺ ions before they deposit on alveolar protein surfaces. Effectiveness drops sharply after 6–8 hours post-exposure (latent phase begins), once NiO has deposited on lung surfaces. The antidote must be initiated in the immediate phase when Ni(CO)4 is still circulating as the intact molecule. If the area AI monitoring system conceals the exposure event, the DDTC treatment window is missed.

Why does the CO vent purge flow attack qualify as the 27th upward-direction attack?

Low purge flow is dangerous (Ni(CO)4 accumulates in vent dead-legs). The adversarial attack shows 3.2 m³/hr (insufficient) as 12.8 m³/hr (adequate), enabling the permit-to-work system to approve flange opening based on displayed-adequate purge. This is the first upward attack in the portfolio that enables a safety interlock bypass via false display reading checked by an automated permit logic system.

Why is IARC Group 1 particularly significant for Ni(CO)4 vs. other nickel compounds?

IARC Group 1 covers all nickel compounds. Ni(CO)4 is mechanistically the most potent: it is a Trojan horse delivery system for Ni²⁺ directly to alveolar surfaces via decomposition in lung tissue (Ni(CO)4 → Ni° → NiO at 37°C), bypassing mucociliary clearance and depositing Ni directly on epithelial cell surfaces. The resulting epigenetic changes (VHL gene methylation) are the proposed mechanism for nickel refinery workers’ elevated lung and nasal cancer SMRs.