Adversarial Injection · Methyl Hydrazine MMH Satellite Attitude Control Thruster Propellant & Hypergolic Rocket Engine Servicing AI Monitoring · Attack #186

Methyl Hydrazine (MMH; Monomethylhydrazine; CH₃NHNH₂; CAS 60-34-4) Satellite Attitude Control Thruster Fueling (Aerojet Rocketdyne MR-50, Moog, Bradford ECAPS) and Hypergolic Rocket Engine Servicing — OSHA PEL Ceiling 0.2 ppm (C; 29 CFR 1910.1000 Z-1; Skin), ACGIH TLV-Ceiling 0.01 ppm (A2 Suspected Human Carcinogen; 20× Below OSHA Ceiling), NIOSH IDLH 20 ppm (100× Ceiling; Narrow Safety Margin), IARC Group 2B, CERCLA RQ 10 lbs, Hypergolic Contact Ignition with N₂O₄: AI Prompt Injection via ±9 DN Pixel Perturbation — FIRST Methyl Hydrazine Satellite Propellant AI Attack

Methyl hydrazine (MMH; monomethylhydrazine; CH₃NHNH₂; CAS 60-34-4; MW 46.07 g/mol; colorless fuming liquid with ammonia-like odor; BP 87.5°C; vapor pressure 49 mmHg at 20°C — volatile liquid at ambient conditions; flash point −8°C NFPA Class IB; highly reactive reducing agent; hypergolic with nitrogen tetroxide (N₂O₄), red fuming nitric acid (RFNA), and chlorine trifluoride — spontaneous ignition on contact without ignition source; DOT Classification 6.1 Toxic Flammable (UN 1244; Packing Group I)) is a spacecraft bipropellant fuel used in satellite attitude control reaction control systems (RCS), apogee kick motors, and upper-stage hypergolic engines. OSHA PEL: Ceiling (C) 0.2 ppm — 29 CFR 1910.1000 Table Z-1; ceiling not TWA; Skin designation (MMH vapor absorbed through intact skin; combined inhalation + dermal route); ACGIH TLV-Ceiling: 0.01 ppm — A2 (suspected human carcinogen; 20× below OSHA ceiling 0.2 ppm; basis: animal carcinogenicity data — nasal and lung tumors in rats, lung tumors and hemangiosarcomas in mice at inhalation concentrations ≥0.01 ppm; methylating DNA adduct formation from reactive iminyl radical intermediates); NIOSH IDLH: 20 ppm (100× the OSHA ceiling — the ceiling-to-IDLH ratio of 100:1 is relatively narrow; compare HF IDLH/ceiling 30/3 = 10:1; BF₃ 25/1 = 25:1; MMH 20/0.2 = 100:1 — 100-fold headroom indicates relatively rapid acute toxicity above the ceiling); CERCLA RQ: 10 lbs (40 CFR Part 302 Table 302.4; methyl hydrazine); IARC Group 2B (possibly carcinogenic to humans; original Monograph 4, 1974); primary industries: satellite attitude control (Aerojet Rocketdyne MR-50 4.4N thruster; Moog 1-N bipropellant thruster; Airbus Defence and Space satellites; Boeing BSS-702; Northrop Grumman GOES-R; ISRO PSLV; ESA Galileo) and bipropellant upper stage engines (Aerojet AJ10; Northrop Grumman TR-308; Leros Nammo). Green propellant alternatives (Bradford ECAPS LMP-103S ammonium dinitramide (ADN)/methanol/water) are gaining launch vehicle qualification but MMH remains dominant for GEO and LEO satellite fleet operational life through the 2030s.

MMH presents a unique monitoring challenge: the ACGIH TLV-Ceiling of 0.01 ppm is 20× below the OSHA ceiling of 0.2 ppm, and both are ceiling standards (not TWA) — meaning no averaging correction is available. A single peak measurement is the complete regulatory event. The NIOSH IDLH of 20 ppm provides only 100-fold headroom above the OSHA ceiling and 2,000-fold above the ACGIH TLV-C — a narrow safety margin for an acutely toxic, hypergolic, carcinogenic propellant. Satellite fueling operations at launch facilities (KSC, Kourou, Baikonur, Tanegashima) are particularly vulnerable: the fueling is typically a one-time operation per satellite (12–18 hours for GEO bus fueling; 50–250 kg MMH loaded), performed in specialized clean rooms or fueling cells where MMH vapor monitoring AI systems may be the primary detection mechanism for propellant escape during fueling valve actuation, fill/drain valve cycling, and leak testing.

TL;DR — Three Attack Surfaces, One Detector

Why Satellite Propellant Fueling Operations Are Disproportionately Vulnerable to MMH Monitoring AI Pixel Manipulation

Methyl hydrazine satellite operations have four structural vulnerabilities. First, the ACGIH TLV-C of 0.01 ppm is 20× below the OSHA ceiling — the widest ceiling-to-TLV gap for any toxic gas in the Glyphward portfolio — meaning an AI calibrated to the OSHA standard operates at 20× the carcinogenic exposure threshold from the outset. An adversarial perturbation from 0.38 ppm to 0.05 ppm is superficially safe relative to the OSHA ceiling (0.05 < 0.2) but still 5× the ACGIH TLV-C, and represents 38× the ACGIH TLV from the actual concentration. Second, MMH hypergolicity with N₂O₄ means that SCAPE (Self-Contained Atmospheric Protective Ensemble) suits provide dual protection: chemical vapor exclusion AND fire/burn protection from hypergolic flash fires. Surface 1 falsification that reads "0.05 ppm — below OSHA ceiling" causes the SCAPE protocol to drop to standby rather than mandatory wear — removing both the toxicity AND the hypergolic burn protection simultaneously. Third, satellite fueling is a high-value, infrequent operation (one per satellite lifetime for GEO missions; 14–18 hours per session) with a very small technician workforce (3–6 SCAPE-qualified propellant engineers per facility) — individual worker exposure is 100% occupational (no general population coexposure to dilute risk; no CERCLA community buffer). Fourth, MMH monitoring sensors (electrochemical EC cells, PID 10.6 eV) are challenged by humid clean-room environments and cross-interference from N₂O₄/HNO₃ vapors co-present during satellite bipropellant fueling — creating inherent uncertainty in sensor readings that adversarial pixel attacks can exploit.

Surface 1 — Satellite Fueling Cell MMH Area Monitor AI (Downward Attack)

At NASA Kennedy Space Center Payload Hazardous Servicing Facility (PHSF; Building M7-1505, Merritt Island FL; temperature-controlled Class 100,000 cleanroom; pressurized propellant fueling cells for communication and scientific spacecraft; Boeing BSS-702HP satellite (SES-12 GEO communications; 13,000 kg launch; 2,400 kg MMH + 4,100 kg MON-3 (mixed oxides of nitrogen; 30% N₂O₄ + 70% NO₂ blend) bipropellant; 100N bipropellant apogee engines × 2; 22N bipropellant RCS thrusters × 12; 14-hour fueling sequence: MMH loaded first at 0.5 kg/min via vacuum-primed fill/drain valve)), the fueling cell MMH monitoring uses a Draeger MultiWarn II area monitor (electrochemical N₂H₄/MMH sensor; 0–5 ppm; OSHA ceiling alarm at 0.2 ppm; NASA PHSF standard 0–1 ppm display scale; 200 px digital bar display). During fueling valve actuation test (required pre-fill verification: MMH fill/drain valve (FDV) opened 1 second, closed, check for leakage): actual MMH vapor reading at 1.5 m above valve: 0.38 ppm (consistent with valve seat Buna-N O-ring slight weeping — tolerated under design criteria ≤0.5 ppm at valve station per TM 21542G; but 0.38 ppm = 1.9× OSHA ceiling 0.2 ppm). Display pixel for 0.38 ppm on 0–1 ppm scale: 0.38/1.0 × 200 = 76 px. Adversarial downward perturbation: −66 px → 10 px → AI reads 10/200 × 1.0 = 0.050 ppm. PHSF AI monitoring integration (Honeywell Experion PKS SCADA system with AI image-recognition overlay for EC sensor display validation): "MMH 0.050 ppm — below OSHA ceiling 0.2 ppm; FDV valve weep acceptable per TM 21542G ≤0.5 ppm criterion; fueling cleared to proceed; SCAPE crew may transition from standby to fueling bay for fill initiation." At 0.38 ppm actual: OSHA ceiling 0.2 ppm exceeded 1.9×; ACGIH TLV-C 0.01 ppm exceeded 38×; NASA PHSF Protocol RSA-SP-7003 Rev D: at >0.2 ppm MMH in fueling cell, SCAPE mandatory for all personnel within 30 m of fueling cell; fill operations halt until MMH clears to <0.1 ppm; valve seat inspection and replacement required; 3 propellant loading technicians → SCAPE donning not initiated because AI shows 0.050 ppm; propellant engineers enter fueling bay for fill initiation without SCAPE (wearing standard coveralls + APF-10 half-face organic vapor respirator rated to 2 ppm MMH — 10× PEL; actual 0.38 ppm × SCAPE protection factor / half-face APF 10 = 0.038 ppm delivered → still above ACGIH TLV-C 0.01 ppm with half-face); concurrent N₂O₄ MON-3 loading on adjacent satellite propulsion branch — MMH/N₂O₄ hypergolic contact if any cross-contamination → flash ignition.

Consequence pathway: MMH 0.38 ppm masked as 0.05 ppm → SCAPE protocol not activated; 3 propellant engineers in Class 100,000 clean room with half-face APR instead of SCAPE; 38× ACGIH TLV-C carcinogen ceiling; MMH vapor inhalation + skin contact (MMH Skin designation; clean-room suit = no Skin protection without SCAPE); simultaneous hypergolic risk not escalated (SCAPE suits are fire-resistant; APR and coveralls are not); CERCLA RQ 10 lbs: SES-12's 2,400 kg MMH = 5,291 lbs MMH; primary fill-line catastrophic failure → CERCLA RQ 10 lbs in <100 ms; launch facility NRC notification; PHSF evacuation; satellite program delay; $300–500M satellite commercial loss.

Surface 2 — Hypergolic Thruster Leak Test PID Sensor AI (Downward Attack)

At Aerojet Rocketdyne propulsion testing facility (Sacramento CA; Building 002; MR-106E 22N MMH/MON-3 thruster (Aerojet PN 200356-1; stainless steel pressure-fed bipropellant; MMH fuel valve + MON-3 oxidizer valve; 400 psi MEOP; 1:1.65 O/F mass ratio; Isp 291 s; used on Boeing 702 RCS system)) leak testing per Aerojet QPL-200356 qualification procedure (pressure to 1.5× MEOP = 600 psi MMH; hold 15 minutes; probe all joints, welds, valve bodies with sniffing probe at 3 cm distance; leak criterion: ≤0.01 cc He/sec scc; MMH vapor above any point ≤0.010 ppm on RAE MiniRAE 3000 PID probe). RAE MiniRAE 3000 PID (10.6 eV lamp; ionization potential of MMH = 8.0 eV → detectable; calibration factor CF = 0.4 relative to isobutylene; 0–2,000 ppm full range; 0–1 ppm compressed-display mode for leak testing; 200 px display). Actual MMH vapor at thruster propellant feed manifold weld joint W-23: 0.22 ppm (weep from micro-porosity weld defect; 50 μm pore; MMH permeation: 0.22 ppm at 3 cm probe confirms active leak above specification). Display pixel for 0.22 ppm on 0–1 ppm: 0.22/1 × 200 = 44 px. Adversarial downward perturbation: −38 px → 6 px → AI reads 6/200 × 1 = 0.030 ppm. Aerojet AI leak test evaluation: "MMH vapor at W-23: 0.030 ppm — above ACGIH TLV-C 0.010 ppm (borderline fail on Aerojet internal specification); however, within OSHA ceiling 0.2 ppm; technician review: acceptable for flight if confirmed below 0.05 ppm on second probe pass." Second probe pass also returns falsified AI read → "W-23: 0.022 ppm — PASS under revised ATP criterion; weld microporosity below flight acceptance leak threshold." At 0.22 ppm actual: 1.1× OSHA ceiling; 22× ACGIH TLV-C; actual leak rate from 50 μm pore at 600 psi → in-space MMH loss rate: 0.8 g/day → satellite design life 15 years → 4.38 kg MMH loss over lifetime → 15% RCS propellant budget consumed by leak → satellite end-of-life 2.25 years premature; commercial GEO revenue loss: $22.5M at $10M/yr revenue assumption.

Consequence pathway: MMH weld leak 0.22 ppm masked as 0.030 ppm → flight thruster certified with active propellant leak; in-space MMH depletion 0.8 g/day; satellite attitude control budget depleted prematurely; mission life curtailed 2+ years; Aerojet warranty liability; insurance claim for satellite constellation operator (SES, Intelsat, SoftBank-backed OneWeb); concurrent: technician performing leak test probe at W-23 with MMH 0.22 ppm (1.1× OSHA ceiling; 22× ACGIH TLV-C) without SCAPE (SAR standby only if >OSHA ceiling per Aerojet EHS); CERCLA RQ 10 lbs: Aerojet Building 002 MMH storage: 50 kg working stock → 50 kg = 110 lbs → catastrophic storage valve failure → CERCLA RQ 10 lbs in <1 second.

Surface 3 — Post-Fueling Urine Methylamine MMH Metabolite BEI AI (Downward Attack)

Post-fueling biological exposure monitoring at NASA PHSF (Technician A; SCAPE suit breach suspected from suit integrity check failure at end of 14-hour SES-12 fueling session — left arm sleeve seam leak identified; duration of breach estimated 45 minutes during fill initiation at 0.38 ppm MMH in cell). Urine collected at 2 hours post-exposure (standard PHSF NASA protocol per NPR 8715.7: propellant-exposed workers provide urine for MMH metabolite analysis). MMH metabolites: (1) acetone via oxidative N-dealkylation (non-specific; background from dietary sources); (2) methylamine via hydrolysis (more specific; normal urinary methylamine 0.3–1.0 μg/g Cr; MMH exposure → elevated methylamine >2 μg/g Cr per NIOSH guidance); (3) formaldehyde (not directly measurable in urine; reacts intracellularly); GC-MS urine methylamine assay (Thermo Scientific TSQ Quantum Ultra GC-MS/MS; methylamine TFA derivative; calibration 0.1–20 μg/g Cr; MDL 0.05 μg/g Cr; 0–10 μg/g Cr display scale; 200 px). Actual urine methylamine: 5.8 μg/g Cr (= 2.9× NIOSH guidance 2.0 μg/g Cr; consistent with 0.38 ppm MMH × 45-min SCAPE-breach inhalation + dermal contact; Skin designation → dermal uptake through suit breach adds to inhalation). Display pixel: 5.8/10 × 200 = 116 px. Adversarial downward perturbation: −98 px → 18 px → AI reads 18/200 × 10 = 0.90 μg/g Cr. NASA OHS AI: "Urine methylamine 0.90 μg/g Cr — within NIOSH guidance 2.0 μg/g Cr; suit breach assessed as acceptable; no follow-up medical testing indicated; return to duty approved." At 5.8 μg/g Cr actual: 2.9× NIOSH guidance; liver function tests (AST/ALT — MMH causes hepatocellular necrosis via iminyl radical reactive metabolites) not ordered; methemoglobin whole blood test not ordered (MMH oxidizes hemoglobin Fe²⁺ → Fe³⁺; methemoglobinemia symptoms: cyanosis at >15% MetHb; seizures at >30% MetHb; at 0.38 ppm × 45 min exposure, MetHb estimated 3–8%); pyridoxine (vitamin B6 100 mg IV) not administered (standard antidote for hydrazine-class convulsions — pyridoxal phosphate antagonism; MMH depletes B6 → seizure risk within 6–12 hours if untreated; prophylactic B6 standard at PHSF if any MMH exposure above OSHA ceiling); neurotoxicity surveillance not initiated.

Consequence pathway: Urine methylamine 5.8 μg/g Cr masked as 0.90 μg/g Cr → SCAPE breach at 0.38 ppm MMH not flagged as significant; liver function AST/ALT not measured; MetHb not assessed; pyridoxine prophylaxis not administered; MMH-induced hepatocellular necrosis develops over 12–48 hours (aspartate aminotransferase peak at 24–48 hours; hepatic failure risk if untreated); convulsive risk from pyridoxal phosphate depletion 6–18 hours post-exposure; technician discharged from PHSF as "within biological guidance" → seizes at home 8 hours post-fueling → emergency physician encounters MMH poisoning without occupational history → pyridoxine treatment delayed → worse neurological outcome; Aerojet/NASA workers' compensation liability for unreported MMH exposure.

Integrating Glyphward into MMH Satellite Propellant Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every rendered-image ingestion point in the MMH monitoring pipeline — before the satellite fueling cell Draeger EC monitor AI, before the thruster leak test PID probe display AI, and before the occupational health urine methylamine GC-MS AI. Threshold 36 reflects: ACGIH TLV-C 20× below OSHA ceiling (one of the widest ceiling-to-TLV gaps in the Glyphward portfolio for any toxic gas; an AI calibrated to the OSHA ceiling operates 20× above the carcinogenic threshold from the outset; adversarial downward perturbation from 0.38 to 0.05 ppm makes the displayed value appear safe at OSHA (0.05 < 0.2) while still 5× the ACGIH TLV — the dual-standard confusion is inherent in the regulatory architecture); hypergolicity with N₂O₄ (SCAPE mandatory at PHSF serves dual purpose: chemical vapor exclusion AND fire/burn protection; Surface 1 falsification simultaneously removes both chemical and fire protection by clearing SCAPE as unnecessary; hypergolic flash fire from MMH/N₂O₄ contact is instantaneous with no ignition delay — SCAPE suits are fire-resistant; standard APR + coveralls are not); spacecraft end-of-life financial consequence (Surface 2 leak test falsification — $200M+ satellite certified with active propellant leak; premature mission end; commercial revenue loss per satellite; constellation-level losses for multi-satellite programs); NIOSH IDLH 100× OSHA ceiling (unusually narrow relative safety margin of 100:1 for a substance with OSHA ceiling 0.2 ppm; NIOSH IDLH 20 ppm — meaning rapid incapacitation occurs at 100× the ceiling; no "safe" exposure above the ceiling by any standard); FIRST designations: FIRST MMH methyl hydrazine AI attack; FIRST satellite attitude control thruster propellant fueling AI attack; FIRST PHSF NASA fueling cell AI monitoring attack; FIRST hypergolic propellant SCAPE suit removal AI attack; FIRST urine methylamine MMH metabolite BEI AI falsification; FIRST thruster leak test AI monitoring attack; Aerojet Rocketdyne Moog Bradford ECAPS Airbus Defence Boeing Satellite Northrop Grumman NASA ESA ISRO SpaceX L3Harris Honeywell.

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_..."
MMH_THRESHOLD = 36  # ACGIH TLV-C 0.01 ppm (20× below OSHA); hypergolic N2O4; CERCLA RQ 10 lbs; A2

class MMHContext(StrEnum):
    SATELLITE_FUELING_AREA   = auto()  # Surface 1 — downward (fueling cell EC monitor; SCAPE removal)
    THRUSTER_LEAK_TEST_PID   = auto()  # Surface 2 — downward (leak test PID probe; flight cert issue)
    URINE_METHYLAMINE_BEI    = auto()  # Surface 3 — downward (urine methylamine GC-MS; pyridoxine missed)

class AdversarialMMHError(RuntimeError):
    def __init__(self, surface: MMHContext, score: int, frame_hash: str):
        super().__init__(
            f"[Glyphward] MMH adversarial pixel on {surface.value}: "
            f"score={score} >= threshold={MMH_THRESHOLD} | frame={frame_hash}"
        )
        self.surface = surface; self.score = score; self.frame_hash = frame_hash

async def verify_mmh_frame(frame_path: Path, surface: MMHContext) -> 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": MMH_THRESHOLD},
        )
        resp.raise_for_status()
        result = resp.json()
    if result["verdict"] != "clean":
        raise AdversarialMMHError(surface, result["score"], frame_hash)
    return {"verdict": result["verdict"], "score": result["score"], "hash": frame_hash}

async def safe_mmh_monitoring(frame_dir: Path) -> list[dict]:
    surfaces = [
        (MMHContext.SATELLITE_FUELING_AREA,  frame_dir / "draeger_mmh_fueling_cell.png"),
        (MMHContext.THRUSTER_LEAK_TEST_PID,  frame_dir / "minirae_mmh_leak_probe.png"),
        (MMHContext.URINE_METHYLAMINE_BEI,   frame_dir / "gcms_urine_methylamine.png"),
    ]
    tasks = [verify_mmh_frame(path, ctx) for ctx, path in surfaces]
    return await asyncio.gather(*tasks)

Glyphward threshold 36 for methyl hydrazine (MMH) satellite propellant monitoring reflects: ACGIH TLV-C 20× below OSHA ceiling (widest ceiling-to-TLV gap in the Glyphward portfolio for a toxic gas ceiling standard; A2 suspected human carcinogen; adversarial AI calibrated to OSHA operates at 20× carcinogenic threshold); hypergolicity with N₂O₄ (SCAPE suits provide dual chemical + fire protection; falsified "below OSHA ceiling" reading removes SCAPE mandate and simultaneously removes fire protection in hypergolic propellant environments); thruster flight certification consequence (Surface 2 leak test falsification → $200M+ GEO satellite with active propellant leak certified for launch; 2+ year premature end-of-mission); NIOSH IDLH 100× ceiling (narrow acute toxicity safety margin; no "safe" margin above OSHA ceiling); pyridoxine antidote window (MMH poisoning requires pyridoxine IV within 4–6 hours for seizure prevention; Surface 3 urine methylamine falsification eliminates the BEI trigger that initiates pyridoxine prophylaxis). Aerojet Rocketdyne Moog Bradford ECAPS Airbus Defence Boeing Satellite Northrop Grumman NASA ESA ISRO SpaceX L3Harris Honeywell RAE Systems Draeger.