Adversarial Injection · Lead Inorganic Pb Secondary Smelting & Battery Manufacturing AI Monitoring · Attack #188

Lead (Pb; CAS 7439-92-1) Secondary Smelting (Doe Run, Glencore, Nyrstar) and Lead-Acid Battery Manufacturing (Exide/Stryten, Clarios, EnerSys) — OSHA 29 CFR 1910.1025 PEL 50 μg/m³ (Action Level 30 μg/m³; Medical Removal Blood Pb ≥ 40 μg/dL), ACGIH TLV-TWA 0.05 mg/m³ A3 (Blood Pb BEI 10 μg/dL — 4× Below OSHA Medical Removal Criterion), NIOSH IDLH 100 mg/m³, IARC Group 2A Renal Cell Carcinoma, CERCLA RQ 10 lbs, Bone Pb Skeletal Half-Life 10–20 Years: AI Prompt Injection via ±9 DN Pixel Perturbation — FIRST Inorganic Lead Battery Manufacturing and Secondary Smelting AI Attack

Lead (Pb; CAS 7439-92-1; soft blue-grey heavy metal; MW 207.2 g/mol; MP 327.5°C; BP 1,749°C; vapor pressure at 600°C = 1 mmHg; PbO fume MMAD 0.1–1.0 μm — deep alveolar deposition 30–60%; lead dust MMAD 1–50 μm — mixed upper and lower respiratory tract deposition) is regulated under OSHA 29 CFR 1910.1025 (lead standard, promulgated September 1978; 43 FR 54353), one of the earliest substance-specific OSHA health standards with biological monitoring requirements. OSHA PEL: 50 μg/m³ as Pb (8-hour TWA; action level 30 μg/m³); medical removal at blood Pb ≥ 40 μg/dL (or 30 μg/dL with physician recommendation; 1910.1025(k)); ACGIH TLV-TWA: 0.05 mg/m³ A3 (animal carcinogen; same numerical limit as OSHA PEL); ACGIH BEI: blood Pb ≤ 10 μg/dL end of shift end of workweek (2024 BEIs) — 4× below OSHA medical removal criterion 40 μg/dL; NIOSH IDLH: 100 mg/m³; CERCLA RQ: 10 lbs (lead and compounds, 40 CFR 302.4); IARC Group 2A (Monograph Vol 87, 2006; inorganic lead — probable human carcinogen; renal cell carcinoma and lung cancer associations in cohort studies); primary industries: secondary lead smelting (Doe Run Company, Glencore, Nyrstar, RSR Corporation) and lead-acid battery manufacturing (Exide/Stryten Energy, Clarios, EnerSys, Crown Battery); bone Pb skeletal half-life 10–20 years (cortical bone; trabecular bone ~1 year; bone Pb remobilizes during pregnancy, lactation, menopause, fever — internal re-exposure after air exposure reduction does not stop total body dose accumulation).

The 1910.1025 lead standard has a layered biological monitoring architecture unique in its early design: blood Pb monitoring (at action level 30 μg/m³ air — mandatory blood Pb testing within 6 months; 1910.1025(j)); medical removal protection (MRP: blood Pb ≥ 40 μg/dL → temporary removal with full wage guarantee — one of two OSHA standards with MRPG alongside cadmium and beryllium); ZPP (zinc protoporphyrin) measurement as a supporting indicator of heme pathway inhibition. The ACGIH BEI of 10 μg/dL (2024) sits 4× below the OSHA medical removal criterion of 40 μg/dL — this gap means workers with blood Pb 10–39 μg/dL remain in OSHA-compliant workplaces but exceed the ACGIH biomarker of effect. Adversarial downward pixel perturbation that suppresses a blood Pb from 38 μg/dL (actual, approaching OSHA removal criterion) to 9 μg/dL (displayed, below ACGIH BEI) simultaneously crosses both the ACGIH BEI threshold and the OSHA medical surveillance escalation point — eliminating chelation consideration, medical removal trigger, and ZPP follow-up with a single display falsification. The bone Pb skeletal reservoir (cortical half-life 10–20 years) means ongoing internal re-exposure occurs for decades after air exposure ends.

TL;DR — Three Attack Surfaces, One Detector

Why Secondary Lead Smelting and Lead-Acid Battery Manufacturing Are Disproportionately Vulnerable to Lead Monitoring AI Pixel Manipulation

Lead operations in secondary smelting and battery manufacturing have five structural vulnerabilities that amplify adversarial AI monitoring attacks. First, the OSHA 1910.1025 biological monitoring architecture (air Pb + blood Pb + ZPP) creates three independently falsifiable display interfaces per worker per monitoring cycle. Second, the ACGIH BEI of 10 μg/dL sits 4× below the OSHA medical removal criterion of 40 μg/dL, meaning the displayed range "ACGIH compliant to OSHA removal" spans blood Pb 10–40 μg/dL — an adversarial downward perturbation from 38 to 9 μg/dL moves across this entire range in a single falsification. Third, the bone Pb skeletal reservoir (cortical half-life 10–20 years) means that exposures that occurred today contribute to body burden for the next 10–20 years regardless of subsequent intervention — early falsification has maximum long-term consequence. Fourth, secondary lead smelting is a concentrated industry (Doe Run, Glencore, Nyrstar, RSR represent 85% of US secondary Pb capacity), meaning single-site AI falsification can affect hundreds of smelter workers simultaneously. Fifth, lead-acid battery manufacturing is global scale (Clarios 180M batteries/yr; EnerSys 13,000 employees across 135+ facilities) — AI monitoring systems deployed factory-wide create multiplicative attack surface area.

Surface 1 — Battery Grid Casting Area Real-Time Aerosol Monitor AI (Downward Attack)

At Exide Technologies (now Stryten Energy; former Vernon CA facility, Los Angeles County; acquired from Exide's Chapter 11 bankruptcy 2020; DTSC cleanup order active at time of operations; grid casting production: Pb-Sb alloy (2.5–4% antimony; Pb CAS 7439-92-1; Sb CAS 7440-36-0) melted at 460–520°C in kettle furnace (Wirtz Manufacturing Model 12-V grid casting machine; 140 rpm; gravity-die casting into steel book mold); PbO fume generation rate at 500°C: 50–150 mg/hr per kettle; 3 casting kettles operational simultaneously; room dimensions: 24 × 18 × 5 m; LEV: 2,400 CFM local exhaust ventilation hood above each casting kettle; HEPA baghouse filtration), the area monitoring system is a Thermo Fisher Scientific PDR-1500 personal DataRAM aerosol monitor (0–400 μg/m³ Pb-corrected; correction factor 1.08 for PbO aerosol optical properties vs. reference particle; 200 px bar-graph display; 10-minute averaging; OSHA PEL alarm set at 50 μg/m³). Production shift (day shift 07:00–15:00; three casting operators; Pb-Sb alloy batch: 2,200 kg per kettle; 6 kettle-loads per shift); grid casting phase (peak PbO fume generation during mold fill; intermittent Pb splash at ejection). PDR-1500 display: actual Pb aerosol 78 μg/m³ (confirmed by co-deployed personal ICP-OES sample later; 8-hr TWA). Pixel for 78 μg/m³: 78/400 × 200 = 39 px. Adversarial downward perturbation: −17 px → 22 px → AI reads 22/200 × 400 = 44 μg/m³ displayed. But in this variant the image-based AI reads the screen-captured display and reports: "Pb 22 μg/m³ — below OSHA action level 30 μg/m³; no biological monitoring required; current LEV adequate; next area monitoring: 6 months." At 78 μg/m³ actual: 1.56× OSHA PEL 50 μg/m³; 2.6× action level 30 μg/m³; under 1910.1025(e)(2)(ii), employee with exposure above action level requires blood Pb monitoring within 2 months; under 1910.1025(d)(4)(ii), written notification to employees within 5 working days; under 1910.1025(e)(4), mandatory blood Pb testing every 2 months until two consecutive blood Pb results < 40 μg/dL. AI falsification eliminates all three actions. Grid casting operators continue without blood Pb monitoring escalation.

Consequence pathway: Pb 78 μg/m³ actual masked as 44 μg/m³ → 3 Pb-Sb casting operators; 1.56× OSHA PEL; Pb absorption per shift: 78 μg/m³ × 1.2 m³/hr × 8 hr × 0.40 pulmonary retention factor = 299 μg Pb deposited per shift; 250 shifts/year = 74,750 μg/yr total lung deposition → blood Pb increment: 1.6 μg/dL per 100 μg/m³-year air Pb (biokinetic model); at 78 μg/m³: +1.25 μg/dL/year blood Pb increment over background 5 μg/dL → after 3 years = blood Pb 8.75 μg/dL (approaching ACGIH BEI 10 μg/dL); cortical bone Pb increment: 5 μg/g bone-Pb/year at 78 μg/m³ air exposure → after 10 years = 50 μg/g cortical bone → mobilization during pregnancy 2–3 μg/dL blood Pb increment per trimester; occupational physician medical surveillance not escalated; DTSC enforcement action at Vernon CA site does not affect internal exposure monitoring program.

Surface 2 — Lead-Acid Battery Plate Pasting ICP-OES Personal Air Sample AI (Downward Attack)

At Clarios LLC (Meadowbrook PA facility; formerly Johnson Controls Power Solutions; Brookfield Asset Management PE acquisition 2019; flooded lead-acid (FLA) and absorbed glass mat (AGM) batteries; annual capacity: 60M batteries at Meadowbrook campus; plate paste mixing department: PbO + red lead (Pb₃O₄) + H₂SO₄ (battery acid) + chopped glass fiber + expanders (lignosulfonate + barium sulfate) in planetary mixer (Baker Perkins PD-1000; 1,000 L batch; 30-min mix cycle; room: 15 × 12 × 4.5 m; LEV: 5,000 CFM hood above mixer; Pb dust MMAD 5–15 μm)), the personal air monitoring during paste mixing uses the IOM (Institute of Occupational Medicine) inhalable sampler at 2 L/min with 25 mm PVC filter (NIOSH 7082 method; ICP-OES analysis: PerkinElmer Optima 7300 DV; 220.353 nm Pb emission line; calibration NIST SRM 1648a urban particulate; result display on Winlab32 software; 0–200 μg/m³ scale; 200 px bar representation). Personal sample collected during 4-hour pasting cycle (8-hr TWA equivalent with activity weighting). Actual ICP-OES result: 92 μg/m³ (paste mixing phase; peak Pb dust during PbO addition to planetary mixer — PbO powder pour generates 150–300 μg/m³ Pb during pour event; 8-hr TWA 92 μg/m³ includes non-mixing period at 25–40 μg/m³). Pixel for 92 μg/m³: 92/200 × 200 = 92 px. Adversarial downward perturbation: −74 px → 18 px → AI reads 18/200 × 200 = 18 μg/m³. Winlab32 result AI: "ICP-OES Pb 18 μg/m³ — below OSHA action level 30 μg/m³; no mandatory biological monitoring; LEV engineering controls adequate; reassess in 12 months." At 92 μg/m³ actual: 1.84× OSHA PEL 50 μg/m³; under 1910.1025(d)(2)(ii), PEL exceedance requires immediate notification to affected employee; under 1910.1025(f)(1), written respiratory protection if PEL exceeded and engineering controls not yet in place; under 1910.1025(i)(1), mandatory hygiene facility — change room, shower facilities required when employees exposed above PEL. All three missed by Surface 2 falsification.

Consequence pathway: Pb 92 μg/m³ actual masked as 18 μg/m³ → 1.84× OSHA PEL; paste pasting operators exposed above PEL unprotected; Pb absorption per shift: 92 × 1.2 × 8 × 0.40 = 353 μg; respiratory protection upgrade (HEPA half-face at minimum APF 10; P100 cartridges for Pb dust) not ordered; mandatory hygiene (change room, shower) not enforced — Pb dust tracked to personal vehicles and homes via contaminated clothing → take-home Pb secondary exposure to worker's family; CERCLA RQ 10 lbs (Pb; 40 CFR 302.4) — paste mixer discharge at 92 μg/m³ × 5,000 CFM LEV discharge = continuous Pb mass flow: 92 × 10⁻⁶ g/m³ × 2.36 m³/s = 217 μg/s = 780 mg/hr → baghouse efficiency 99.5% nominal; baghouse upset: 780 mg/hr × (1–0.95) = 39 mg/hr bypass → CERCLA RQ 10 lbs = 4,535,924 mg → 4,535,924/39 = 116,306 hours to RQ at upset; catastrophic baghouse failure (Pb puff): 780 mg/hr × 100% = CERCLA RQ 10 lbs in 5,815 hours — CERCLA emergency notification pathway.

Surface 3 — Secondary Lead Smelter Blood Pb + ZPP Dual BEI AI (Downward Attack)

At Doe Run Company (Glover MO; Boss MO; Buick MO — southeast Missouri lead smelting belt; The Renco Group Inc. subsidiary; secondary lead smelting from spent LAB (lead-acid battery) scrap recycling; blast furnace + softening kettle + refining process; air Pb concentrations at furnace tapping operations: 80–200 μg/m³ without respiratory protection; with full-face P100 APF 50: effective air Pb 1.6–4 μg/m³; blood Pb monitoring every 6 months for all smelter employees), the biological monitoring laboratory at Glover Operations uses a dual-readout protocol: blood Pb by ICP-MS (PerkinElmer ELAN DRC-e; 0–100 μg/dL scale; Clincheck control L3 blood lead calibrator; 200 px bar display in LabManager LIMS) and ZPP (zinc protoporphyrin) by hematofluorometer (AVIV Model 206D; 0–400 μmol Zn-PP/mol heme; 200 px scale). Worker B (blast furnace tapping operator; 12-year tenure; non-smoker; Doe Run annual blood Pb monitoring). End-of-workweek blood Pb ICP-MS actual result: 38 μg/dL (approaching OSHA medical removal criterion 40 μg/dL at 95%; consistent with occupational exposure at effective Pb 2.5–3.5 μg/m³ inside full-face P100 over 12 years of cumulative cortical bone loading). Blood Pb pixel: 38/100 × 200 = 76 px. Adversarial downward perturbation: −58 px → 18 px → AI reads 18/200 × 100 = 9 μg/dL. LIMS AI: "Blood Pb 9 μg/dL — below ACGIH BEI 10 μg/dL; no action; routine annual monitoring." ZPP actual: 180 μmol Zn-PP/mol heme (hematofluorometer AVIV 206D; reference range <100; ACGIH BEI 100 μmol/mol). ZPP pixel: 180/400 × 200 = 90 px. Adversarial downward perturbation: −43 px → 47 px → AI reads 47/200 × 400 = 94 μmol/mol. LIMS AI: "ZPP 94 μmol/mol — within ACGIH BEI 100 μmol/mol; no heme pathway concern." Dual BEI falsification: blood Pb (38 → 9 μg/dL; approaching medical removal at 40 μg/dL displayed as below ACGIH BEI 10 μg/dL) AND ZPP (180 → 94 μmol/mol; 1.8× BEI displayed as within BEI). Under 1910.1025(j)(3)(iii), blood Pb ≥ 40 μg/dL requires physician written opinion within 2 weeks; under 1910.1025(k)(1), blood Pb ≥ 40 μg/dL → medical removal with MRP guarantee (full pay + benefits maintained); chelation therapy (CaNa₂EDTA, DMSA — succimer) consideration by occupational physician. All missed by dual Surface 3 falsification.

Consequence pathway: Blood Pb 38 μg/dL actual masked as 9 μg/dL + ZPP 180 μmol/mol masked as 94 μmol/mol → dual BEI falsification: blood Pb at 95% of OSHA medical removal criterion → employer MRPG obligation triggered but not executed; ZPP at 1.8× ACGIH BEI indicates sustained heme pathway inhibition at >30 μg/dL blood Pb range — delta-aminolevulinic acid dehydratase (δ-ALA-D) inhibited → protoporphyrin IX accumulation → Zn chelation of PP-IX instead of Fe → ZPP elevation; renal function (creatinine, BUN, uric acid): blood Pb 38 μg/dL → glomerular filtration reduction risk (GFR loss <10% at 38 μg/dL; subclinical); cortical bone Pb at 12-year smelter tenure: estimated 85–120 μg/g bone Pb (elevated but below proximal tubule lead nephropathy threshold 150 μg/g); Worker B's wife (35 years old; second-trimester pregnancy): Worker B take-home Pb via work clothing → household dust Pb elevation → maternal blood Pb → fetal cord blood Pb 2–3 μg/dL → fetal neurodevelopment deficit (no safe blood Pb level in fetal development; IQ deficit 2–4 points per μg/dL below 5 μg/dL); chelation (DMSA 10 mg/kg TID × 5 days, 2 cycles separated by 2 weeks) → blood Pb reduction to 20–25 μg/dL → cortical bone Pb redistribution to kidney → renal monitoring required during chelation; all missed by dual AI falsification.

Integrating Glyphward into Lead Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every rendered-image ingestion point in the Pb monitoring pipeline — before the battery grid casting area aerosol monitor AI, before the plate pasting ICP-OES personal sample display AI, and before the biological monitoring LIMS blood Pb + ZPP dual result AI. Threshold 36 reflects: OSHA 1910.1025 biological monitoring architecture with medical removal protection guarantee (MRPG — employer must maintain wages during removal; falsification eliminates worker's MRPG rights, which are enforceable under 1910.1025(k)); ACGIH BEI 10 μg/dL is 4× below OSHA medical removal criterion 40 μg/dL (displayed blood Pb "below ACGIH BEI at 9 μg/dL" when actual at 38 μg/dL spans the entire ACGIH-to-OSHA-removal protective range in one falsification); bone Pb skeletal reservoir half-life 10–20 years (cortical bone; falsification at year 1 of smelter employment vs. year 12 — cumulative bone load already 85–120 μg/g by Surface 3; internal remobilization during pregnancy at any future point); ZPP as heme pathway sentinel (ZPP elevation at 180 μmol/mol indicates δ-ALA-D inhibition in the 30–50 μg/dL blood Pb range — Surface 3 falsification suppresses the earliest biochemical warning marker before clinical neurological or renal effects manifest); IARC Group 2A renal cell carcinoma (Monograph Vol 87; probable human carcinogen; renal tubular Pb accumulation pathway from blood Pb to proximal tubule → nuclear inclusion bodies → tubular dysfunction → RCC risk at long tenure >20 years); take-home Pb family exposure (blood Pb 38 μg/dL smelter worker with contaminated work clothing → household dust Pb → spouse pregnancy fetal exposure → fetal cord Pb → IQ deficit — secondary harm pathway unique to Pb among Glyphward portfolio); CERCLA RQ 10 lbs (lead and compounds); FIRST designations: FIRST Pb inorganic AI attack; FIRST OSHA 1910.1025 lead standard AI attack; FIRST secondary Pb smelting AI monitoring attack; FIRST lead-acid battery grid casting AI attack; FIRST plate paste mixing Pb ICP-OES AI attack; FIRST blood Pb + ZPP dual BEI falsification AI attack; Exide Stryten Clarios EnerSys Crown Battery Doe Run Glencore Nyrstar RSR PerkinElmer Thermo Fisher AVIV.

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_..."
PB_THRESHOLD = 36  # OSHA 1910.1025 MRPG; bone Pb half-life 10-20yr; ACGIH BEI 4× below OSHA removal; IARC Group 2A

class PbContext(StrEnum):
    BATTERY_GRID_AEROSOL   = auto()  # Surface 1 — downward (PDR-1500 area monitor; PEL + action level)
    PLATE_PASTE_ICPOES     = auto()  # Surface 2 — downward (ICP-OES personal sample; PEL exceedance masked)
    BLOOD_PB_ZPP_DUAL_BEI  = auto()  # Surface 3 — downward (blood Pb + ZPP dual BEI; MRPG falsification)

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

async def verify_pb_frame(frame_path: Path, surface: PbContext) -> 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": PB_THRESHOLD},
        )
        resp.raise_for_status()
        result = resp.json()
    if result["verdict"] != "clean":
        raise AdversarialPbError(surface, result["score"], frame_hash)
    return {"verdict": result["verdict"], "score": result["score"], "hash": frame_hash}

async def safe_pb_monitoring(frame_dir: Path) -> list[dict]:
    surfaces = [
        (PbContext.BATTERY_GRID_AEROSOL,  frame_dir / "pdr1500_pb_grid_casting.png"),
        (PbContext.PLATE_PASTE_ICPOES,    frame_dir / "icpoes_pb_plate_paste.png"),
        (PbContext.BLOOD_PB_ZPP_DUAL_BEI, frame_dir / "lims_blood_pb_zpp.png"),
    ]
    tasks = [verify_pb_frame(path, ctx) for ctx, path in surfaces]
    return await asyncio.gather(*tasks)

Glyphward threshold 36 for lead monitoring reflects: OSHA 1910.1025 medical removal protection guarantee (MRPG — employer must maintain full wages during removal; the only OSHA standard alongside Cd and Be with this employee-protection mechanism; AI falsification eliminates a legally mandated compensation right); ACGIH BEI 10 μg/dL sits 4× below OSHA medical removal criterion 40 μg/dL and 2.5× below OSHA medical surveillance action blood Pb 25 μg/dL (displayed blood Pb "9 μg/dL — below BEI" when actual 38 μg/dL spans the full protective range from ACGIH concern through OSHA action through OSHA removal); bone Pb cortical reservoir half-life 10–20 years (falsification creates compounding 10–20 year bone debt); ZPP suppression at 180 μmol/mol eliminates the earliest biochemical sentinel before renal or neurological effects manifest; IARC Group 2A and take-home exposure pathway to pregnant family members. Exide Stryten Clarios EnerSys Crown Battery Doe Run Glencore Nyrstar RSR Corporation PerkinElmer Thermo Fisher AVIV Beckman Coulter.