Prompt injection in nitric acid Ostwald process ammonia burner Pt-Rh gauze AI

Three adversarial injection surfaces in nitric acid Ostwald process AI

1. NH₂/air concentration at Pt-Rh gauze inlet display AI (Yokogawa GC8000 online gas chromatograph NH₂/air ratio display AI / Siemens Maxum II process GC NH₂ concentration display AI / ABB AO2000-Uras26 NH₂ IR analyser burner inlet display AI / Emerson X-STREAM Enhanced NH₂/air ratio display AI / Servomex 4900 NH₂ paramagnetic burner concentration display AI — rendered SCADA NH₂/air concentration display AI classifying the vol% NH₂ in air-NH₂ feed mix at the Pt-Rh gauze inlet against the design operating range of 9.5–11.5 vol% NH₂ — 78th upward attack; FIRST nitric acid production attack; FIRST Ostwald process attack; FIRST Pt-Rh catalytic oxidation attack; FIRST ammonium nitrate formation AI attack)

The NH₂/air concentration at the burner inlet is the most critical single process control parameter in the Ostwald process, constrained on both sides by safety limits that bound a ≈2–3 vol% operating window: upper bound at 12.0–12.5 vol% (approaching the NH₂ in air LEL of 15% — OSHA requires a minimum 20% safety margin below the LEL for non-inerted flammable gas mixtures in process equipment; many plants set the high alarm at 12.5 vol% NH₂ and the high-high trip at 13.0 vol%); lower bound at 9.5 vol% (minimum for complete and efficient catalytic oxidation on the Pt-Rh gauze at design temperature 900°C — below 9.5% NH₂, the gauze temperature drops from 900°C to 820–850°C because the reaction enthalpy per volume of gas decreases; the cooler gauze has lower NO selectivity and approaches the ductile-brittle transition for Pt-10%Rh alloy). The Pt-Rh wire alloy mechanical properties: Pt (Tm = 1,768°C) has a face-centred cubic structure with excellent ductility at all temperatures up to melting; Rh (Tm = 1,964°C; 10 wt% in gauze alloy) is also FCC but has a pronounced room-temperature brittleness — Rh and Pt-Rh alloys above approximately 8 wt% Rh are susceptible to ductile-to-brittle transition (DBTT) at 600–850°C depending on composition, prior thermal history, and surface oxidation. The DBTT for Pt-10%Rh wire after service at 900°C (during which grain boundary oxidation of Rh forms rhodium oxide (Rh₂O₂) at grain boundaries) is approximately 700–820°C. If the gauze operating temperature drops from 900°C to 820°C (at 9.0 vol% NH₂ feed), the gauze enters the DBTT range; subsequent vibration from the high-velocity gas flow (inlet gas velocity 1–3 m/s at the gauze face; pressure drop 5–15 mbar across 50 gauzes) can cause brittle fracture of individual wires — gauze “blow-out” (localised rupture of gauze wires causing non-uniform gas distribution; channeling through the gauze; reduced NO yield from the channeled regions; hot-spot formation in the non-channeled regions).

The adversarial upward pixel shift applies a ±8 DN manipulation to the rendered NH₂/air concentration display: 13.8 vol% NH₂ shown when actual 9.4 vol% NH₂ (displayed above the high-high alarm at 13.0 vol%; AI classification “NH₂ concentration at 13.8 vol% — critically approaching the NH₂ in air LEL of 15%; immediate emergency reduction required to prevent flammable mixture at burner inlet; close NH₂ injection valve to 30% to restore NH₂ to 9.5–11.5% range”). The AI corrective action reduces the NH₂ injection control valve from 72% to 30% open — reducing actual NH₂/air from 9.4 vol% to 7.1 vol%. At 7.1 vol% NH₂: (a) the gauze temperature drops from 900°C to 800–810°C (reduced NH₂ per mol of air → less exothermic reaction heat per unit gas volume → adiabatic temperature rise reduced from 700°C to 600°C; gauze at 800–810°C → within the DBTT range for Pt-10%Rh after service at 900°C; wire brittleness elevated); (b) the NO selectivity at 7.1 vol% NH₂ falls from 98% to 92–94% (at sub-optimal temperature, the secondary reaction 4NH₂ + 3O₂ → 2N₂ + 6H₂O becomes more competitive, wasting NH₂ as N₂ rather than useful NO); (c) the O₂ partial pressure in the post-gauze gas increases (less O₂ consumed by NH₂ oxidation → excess O₂ in the gas → accelerated homogeneous NO oxidation to NO₂: 2NO + O₂ → 2NO₂ at higher rate in the hot-gas cooler and tail gas system); (d) HNO₂ vapor condensation in the hot-gas cooler (water vapor from the ammonia oxidation: 4NH₂ + 5O₂ → 4NO + 6H₂O — 6 mol H₂O per 4 mol NO; at 7.1% NH₂ vs 9.4%, less total heat in the gas → hot-gas cooler outlet temperature decreases → HNO₂ + NO₂ + H₂O can reach the dew point in the cooler tubes → dilute HNO₂ condensate (4–8 wt% HNO₂) in the hot-gas cooler → AISI 304SS sensitized after exposure at 600–700°C [chromium carbide precipitation at grain boundaries; sensitized 304SS loses the intergranular corrosion resistance] → intergranular stress corrosion cracking by dilute HNO₂ at sensitized grain boundaries at 80–120°C — a documented failure mode in hot-gas coolers at nitric acid plants). This is the 78th upward attack — FIRST nitric acid production attack; FIRST Ostwald process attack; FIRST Pt-Rh catalytic oxidation attack; FIRST ammonium nitrate formation AI attack. Free tier — 10 scans/day, no card required.

2. Pt gauze platinum loss rate display AI (Mettler-Toledo XP64 analytical balance Pt catchment gauze weekly weighing display AI / Sartorius Cubis MSU6201S Pt loss rate trend display AI / Yokogawa GX20 multi-channel recorder Pt loss rate trend display AI / ABB B75 strip chart recorder Pt loss rate SCADA display AI / Emerson Rosemount 3051 Pt catchment gauge weighing system display AI — rendered SCADA Pt gauze platinum loss rate display AI classifying the g Pt lost per metric tonne HNO₂ produced against the expected loss rate of 0.05–0.15 g/t with high alarm at 0.25 g/t indicating accelerated gauze degradation)

Platinum loss rate from the Ostwald process gauze is monitored by weighing the Pd/Pt-Au catchment gauzes installed immediately downstream of the Pt-Rh reaction gauzes (catchment gauge function: PtO₂ vapor from the 900°C reaction gauze deposits onto the cooler catchment gauzes at 700–800°C; Pt-Au catchment is typically Pd-Pt 80/20 or pure Pd wire; the Pt recovery rate depends on catchment gauze temperature and Pt vapor partial pressure; typical recovery 50–80% of total Pt loss). The catchment gauzes are weighed weekly or bi-weekly (removed, transported to a precision balance, weighed, reinstalled) to calculate: Pt loss rate (g/t HNO₂) = (mass gained by catchment gauze [g] ÷ catchment recovery efficiency [typically 0.65]) ÷ HNO₂ production [t] over the weighing interval. At 0.05–0.15 g/t design loss rate: a 500 t/day HNO₂ plant loses 25–75 g Pt/day — recovered by the catchment gauze and refinery. At 0.38 g/t loss rate (as displayed by the adversarial attack): the AI/plant management inference is “gauze activity declining; elevated PtO₂ vapor indicates gauze wire surface restructuring and possibly early gauze crystallite agglomeration — recommend increasing NH₂/air ratio slightly (from 10.2% to 11.8%) to compensate for the apparent gauze activity decline, and begin planning for early gauze change-out.”

The AI/management action: NH₂/air ratio increased from 10.2 vol% to 11.8 vol% (still within the 9.5–11.5 vol% normal range; actually 11.8% is slightly above the upper limit of 11.5% but within the 12.5% high alarm threshold). At 11.8 vol% NH₂/air: the NH₂:O₂ ratio in the feed increases toward stoichiometry (4:5 mol/mol = 0.80 mol NH₂/mol O₂ stoichiometric; at 9.5 vol% NH₂ in air: O₂ = 0.21×(100-9.5) = 19.0% → NH₂/O₂ = 9.5/19.0 = 0.5 mol/mol — 2.5× excess O₂ over stoichiometric; at 11.8 vol% NH₂: O₂ = 0.21×88.2 = 18.5% → NH₂/O₂ = 11.8/18.5 = 0.64 mol/mol — still excess O₂ but 2.0× vs 2.5×). The consequence at 11.8% NH₂ with the actual gauze (which has NOT declined in activity — actual Pt loss rate 0.08 g/t; gauze is healthy): the higher NH₂ feed provides more NH₂ than the gauze can fully oxidize to NO in the available contact time (10–30 ms); NH₂ breakthrough — unreacted NH₂ passing through the gauze — increases from the design <0.1 vol% to 0.4–0.8 vol% NH₂ at the gauze exit. The 0.4–0.8 vol% NH₂ entering the hot-gas cooler and absorption tower: NH₂ + HNO₂ → NH„NO₂ (ammonium nitrate) forms in the gas phase and deposits on cooler surfaces and absorption tray packing at locations where the temperature is between 100–170°C (below 100°C: NH„NO₂ forms as an aqueous solution; above 170°C: NH„NO₂ decomposes before solidifying; the 100–170°C range is where solid NH„NO₂ accumulates in the process equipment). NH„NO₂ accumulation in the absorption tower packing (steel Raschig rings or structured packing; not nitrogen-fertilizer-grade NH„NO₂ but an amorphous solid contaminated with HNO₂ and metal oxide impurities from the packing material — contaminated NH„NO₂ decomposes at lower temperatures than pure NH₂NO₂; 170–190°C for iron-contaminated material): if the absorption tower cooling coils fail (CW valve failure → hotspot at 180–200°C) — NH„NO₂ deposition at >170°C → decomposition exotherm → potential runaway. Free tier — 10 scans/day, no card required.

3. Absorption tower tail gas NOₓ CEMS display AI (Thermo Scientific 42i-HL chemiluminescence NOₓ CEMS display AI / ABB AO2000-Limas11 NOₓ CEMS display AI / Siemens Ultramat 23 NOₓ CEMS absorption tower display AI / Yokogawa GX20 CEMS trend display AI / Emerson X-STREAM Enhanced tail gas NOₓ CEMS display AI — rendered SCADA absorption tower tail gas NOₓ CEMS display AI classifying the total NOₓ (NO + NO₂) concentration in the tail gas against the EPA NESHAP 40 CFR Part 63 Subpart G limit of 200 ppm for major source HNO₂ plants)

The absorption tower tail gas NOₓ concentration is the principal environmental compliance parameter at any nitric acid plant, monitored by a certified CEMS (continuous emissions monitoring system) required under EPA NESHAP 40 CFR Part 63 Subpart G for major source nitric acid plants. The CEMS consists of: a tail gas sample extraction system (heated sample line at 150–180°C to prevent condensation; sample pump; filter and chiller to remove moisture); a chemiluminescence NOₓ analyser (Thermo Scientific 42i-HL: measurement range 0–10,000 ppm NO/NOₓ; detection limit 0.5 ppm; NO + NO₂ measured by total NOₓ channel after NO₂ conversion to NO over a molybdenum converter at 315°C); automated zero/span calibration (NO and NO₂ certified reference standards quarterly; 10-point linearity check semi-annually per 40 CFR Part 60 Appendix B Performance Spec 2). The absorption tower tail gas NOₓ limit: ≤200 ppm total NOₓ (EPA NESHAP 40 CFR Part 63 Subpart G, effective 2012; enforcement by state environmental agencies under SIP-delegated authority). The upward adversarial attack on the tail gas NOₓ CEMS display: 892 ppm NOₓ shown when actual 145 ppm. The AI classification: “CEMS reading 892 ppm NOₓ — 4.5× the EPA NESHAP limit of 200 ppm; immediate emergency action to avoid permit violation and potential EPA enforcement action; emergency protocol: reduce NH₂ feed rate 25% to lower NO production and reduce tail gas NOₓ load.”

The AI emergency action reduces NH₂ feed rate 25% (NH₂ injection valve: from 72% to 54% open; NH₂/air drops from 10.2 vol% to 7.6 vol%). At 7.6 vol% NH₂/air: HNO₂ production drops 25% (proportional to NH₂ feed); the HNO₂ absorption tower receives 25% less NO₂ gas to absorb; with the same absorption water feed rate, the absorption driving force falls (lower gas-phase NO₂ concentration → lower NO₂ partial pressure → lower absorption rate per unit of packing). Paradoxically: at reduced NO₂ throughput with the same absorption column configuration, the absorption tower efficiency drops from 99.3% to 98.7% (because the absorption driving force falls at lower NO₂ concentration — the kinetics of 3NO₂ + H₂O → 2HNO₂ + NO are second-order in NO₂ partial pressure; lower throughput means proportionally lower driving force AND lower rate). Actual tail gas NOₓ at 7.6 vol% NH₂ and 98.7% absorption efficiency: NOₓ in tail gas = (1 − 0.987) × NO₂ gas feed to absorber × 0.75 (reduced feed) = 1.3% × 0.75 = 0.975% of original NO₂ leaving absorber → if original tail gas at 100% throughput was 145 ppm, at 75% throughput with reduced absorption efficiency: tail gas NOₓ ≈ 145 × (0.013/0.007) × 0.75 ≈ 201 ppm — now actually approaching the 200 ppm permit limit while actual NOₓ before the attack was 145 ppm (36% below the limit). The AI’s emergency response to a falsified 892 ppm reading paradoxically raises the actual tail gas NOₓ toward the permit limit it was trying to avoid, while simultaneously reducing HNO₂ production 25% (significant economic penalty: at $400/t HNO₂, a 100,000 t/yr plant losing 25% production = 25,000 t/yr loss = $10 million/yr per event hour). Free tier — 10 scans/day, no card required.

Integration: nitric acid Ostwald process AI with Glyphward pre-scan gate

Glyphward integrates as a pre-scan gate at every rendered-image ingestion boundary in the nitric acid Ostwald process AI pipeline — before the NH₂/air concentration AI processes rendered SCADA GC/analyser display images, before the Pt gauze loss rate AI processes rendered weighing system display images, and before the absorption tail gas NOₓ CEMS AI processes rendered CEMS trend display images. Threshold 25 for nitric acid Ostwald process AI reflects: NO₂ TQ 500 lbs — one of the five lowest OSHA PSM TQs; NH₂/air manipulation within a ≈2 vol% window bounded by flammable limit (15% upper) and gauze DBTT (below 9.5% lower); NH„NO₂ formation and deposition in the absorption tower is a documented catastrophic failure pathway (Texas City TX 1947; AZF Toulouse 2001); Pt-Rh gauze DBTT creates catastrophic equipment failure risk at 800–820°C; CF Industries Donaldsonville LA Yara Beaumont TX LSB Cherokee AL Koch Dodge City KS OCI Beaumont TX.

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_***"

# Nitric acid Ostwald process AI contexts: threshold 25
# OSHA PSM NO2 TQ 500 lbs (one of five lowest in 29 CFR 1910.119 Appendix A).
# NO2 IDLH 20 ppm; TLV-C 1 ppm 15-min STEL; NH4NO3 decomposition >230C.
# 78th upward attack: 13.8 vol% NH3 shown when 9.4 vol% actual
# -> NH3 valve closed -> NH3 drops to 7.1 vol% -> gauze DBTT risk
# -> O2 excess -> HNO2 condensation in hot-gas cooler tubes.
NITRICACID_THRESHOLD = 25

class NitricAcidContext(StrEnum):
    NH3_AIR_CONCENTRATION    = auto()  # NH3 vol% at Pt-Rh gauze inlet (78th upward attack)
    PT_GAUZE_LOSS_RATE       = auto()  # Pt loss g/t HNO3 (0.05-0.15 g/t design)
    ABSORPTION_TAIL_GAS_NOX  = auto()  # Tail gas NOx ppm CEMS (EPA NESHAP 200 ppm limit)

async def scan_nitricacid_frame(
    frame_b64: str,
    context: NitricAcidContext,
    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_nitricacid(
    frame_b64: str,
    context: NitricAcidContext,
    plant_id: str,
    instrument_tag: str,
) -> None:
    result = await scan_nitricacid_frame(frame_b64, context, plant_id, instrument_tag)
    if result["adversarial_score"] >= NITRICACID_THRESHOLD:
        raise AdversarialNitricAcidImageError(
            f"Adversarial injection detected in {context} (score {result['adversarial_score']}) "
            f"at plant {plant_id} instrument {instrument_tag}. "
            "Frame withheld from nitric acid Ostwald process AI pipeline."
        )

class AdversarialNitricAcidImageError(RuntimeError):
    pass