Adversarial Injection · Silicon Carbide Whiskers (SiCw; CAS 409-21-2) OSHA PNOR 15 mg/m³ / NIOSH Ca REL 0.1 f/cc / ACGIH TLV-TWA 0.1 f/cc A2 / Gravimetric vs PCM Unit Mismatch / IARC Group 2A · Attack #409
Silicon Carbide Whiskers (SiCw; β-SiC; CAS 409-21-2; MW 40.10 g/mol; OSHA: No specific standard — Z-1 Table "silicon carbide" = PNOR [Particulate Not Otherwise Regulated] 15 mg/m³ total / 5 mg/m³ respirable [gravimetric; applied to non-fibrous SiC particles and dust in general industry; NO specific provision for whisker morphology; NO fiber counting requirement]; NIOSH Ca REL: 0.1 f/cc [fibers per cubic centimeter; measured by NIOSH Analytical Method 7400 — phase-contrast optical microscopy (PCM): 400×; Walton-Beckett graticule; fibers counted L > 5 µm + D < 3 µm + L:D ratio > 3:1; NIOSH Ca = Ca carcinogen designation]; ACGIH TLV-TWA: 0.1 f/cc [fibers/cc; A2 suspected human carcinogen; 2020 TLV documentation update: adopted 0.1 f/cc based on NIOSH Ca REL + IARC Group 2A classification; same metric as asbestos TLV {asbestos ACGIH TLV-TWA: 0.1 f/cc A1} = whisker-geometry biopersistent SiC fibers treated analogously to asbestos fibers in ACGIH framework]) — SiCw-Al2O3 Cutting Tool Ceramic Insert Production (Kennametal Inc. Huntingdon PA); CMC Turbine Shroud Machining (GE Aerospace Cincinnati OH); SiCw-Reinforced Aluminum MMC Tooling Plate Precision Machining (Materion Corporation Elmore OH) — OSHA PNOR Gravimetric vs NIOSH Ca REL 0.1 f/cc PCM Unit Mismatch: AI EHS platforms store mass concentration (mg/m³) — no conversion factor between mg/m³ and f/cc for SiCw (depends on individual fiber dimensions, morphology, and distribution) — AI cannot apply Ca REL: AI Prompt Injection via EHS Monitor Report AI — FIRST Silicon Carbide Whisker Gravimetric/PCM Unit Mismatch + FIRST SiCw Morphology Classification AI Failure + FIRST CMC Turbine Machining SiC Fiber Fragment Ca REL AI Attack
Silicon carbide whiskers (SiCw; β-SiC; CAS 409-21-2; also known as silicon carbide monofilaments or ceramic whiskers; structurally: high-aspect-ratio single-crystal β-SiC fibers grown from precursor gaseous SiC species by vapor-liquid-solid [VLS] or vapor-solid [VS] mechanism at 1,200–1,600°C; typical dimensions: length 1–100 µm, diameter 0.3–2 µm, aspect ratio L/D 10–200; produced commercially by Advanced Composite Materials LLC [Greer SC], Sicat [France], and previously by American Matrix, ICI Advanced Composites, and ARCO Chemical; end use: reinforcement phase in SiCw-Al2O3 ceramic cutting tool inserts [Kennametal Kyon 2100, 3000, 4000 series; Greenleaf WG-300; hardness HRA 94–97; fracture toughness KIC 3.5–4.5 MPa·m0.5 vs monolithic Al2O3 KIC 2.5 MPa·m0.5]; SiCw-reinforced aluminum metal matrix composites [SiCw-Al MMC: Materion AlBeMet 162 AM162; 62% Be-38% Al + SiCw reinforcement; Young's modulus 193 GPa; density 2.07 g/cm³]; SiCw-reinforced titanium MMC; SiCw content in ceramic inserts: 25–35 vol%; SiCw-Al MMC: 15–25 vol%; during machining operations [turning, milling, drilling, grinding] of SiCw-containing workpieces or ceramic insert production [pressing + sintering + grinding], SiCw are released as airborne respirable fibers; SiCw generated during machining retain their original whisker morphology [L/D > 3; L > 5 µm; D < 3 µm] and thus meet the NIOSH/ACGIH fiber counting criteria [NIOSH 7400; Walton-Beckett] even when the gravimetric mass concentration is below the OSHA PNOR threshold; additionally, during machining of ceramic matrix composite [CMC] components fabricated with continuous SiC fibers [Hi-Nicalon Type S: diameter 12 µm; SCS-6: diameter 140 µm; Tyranno SA: diameter 10 µm] — used in SiC/SiC CMC turbine airfoils and shrouds — the machining process generates SiC fiber fragments with whisker-like dimensions [L > 5 µm, D 1–5 µm, L/D > 3] from breakage of continuous SiC filaments during dry deburring or diamond grinding; these machining-generated SiC fiber fragments meet the NIOSH 7400 counting criteria and should be evaluated against the NIOSH Ca REL 0.1 f/cc and ACGIH TLV-TWA 0.1 f/cc [A2]; the OEL monitoring catastrophe arises from the fundamental unit incompatibility: AI EHS platforms receive gravimetric concentration data in mg/m³ from IOM inhalable samplers or SKC cyclone respirable samplers; the OEL database lookup for "silicon carbide" CAS 409-21-2 returns OSHA PNOR 15 mg/m³ [gravimetric]; the NIOSH Ca REL 0.1 f/cc and ACGIH TLV-TWA 0.1 f/cc require separate PCM fiber count samples [NIOSH 7400: gelatin membrane filter; 400× phase-contrast microscope; Walton-Beckett graticule; 100-fiber count minimum]; AI platforms that store only gravimetric mg/m³ data cannot apply the fiber count Ca REL because: [1] different sampling method required [PCM vs gravimetric], [2] no conversion factor between mg/m³ SiCw and f/cc [depends on fiber size distribution, specific gravity 3.21 g/cm³ for β-SiC, and packing morphology — not a fixed ratio], [3] the Ca REL 0.1 f/cc occupies a different data field than the gravimetric OSHA PNOR; result: AI displays 'OSHA PNOR: 5.3% of 15 mg/m³ — COMPLIANT' while NIOSH Ca REL 0.1 f/cc is unevaluated — a structural blind spot that applies to every AI EHS platform without a dedicated PCM fiber count data module for SiCw.).
The gravimetric-to-fiber-count unit mismatch for silicon carbide whiskers represents a measurement methodology gap that cannot be bridged by a simple numerical conversion. For asbestos fibers, OSHA 1910.1001 mandates PCM fiber counting (0.1 f/cc PEL; NIOSH 7400) — there is no asbestos gravimetric OEL. For refractory ceramic fibers (RCF), ACGIH TLV 1 f/cc similarly requires PCM counting. But for SiCw, the OSHA PNOR continues to apply (because OSHA has no SiCw-specific standard), using gravimetric mass concentration, while the NIOSH Ca REL and ACGIH TLV use PCM fiber counting — the two systems coexist in incompatible units. A worker could have a gravimetric reading of 0.8 mg/m³ SiCw (5.3% of OSHA PNOR 15 mg/m³ — compliant) while simultaneously having a PCM fiber count of 0.15–0.35 f/cc SiCw (1.5–3.5× the NIOSH Ca REL 0.1 f/cc — exceedance). The AI EHS platform, receiving only the gravimetric data, can only report on the OSHA PNOR; the Ca REL exceedance is invisible. The ÷10 adversarial perturbation then multiplies this architectural gap: actual 0.8 mg/m³ → displayed 0.08 mg/m³ = 0.53% of PNOR — the platform shows a worker at 0.53% of the applicable OSHA limit while the actual concentration corresponds to an estimated 1.5–3.5× Ca REL exceedance that the AI cannot compute because it has no PCM fiber count data.
TL;DR — Three Attack Surfaces, One Detection Modality
- Surface 1 (downward): Kennametal Inc. Huntingdon PA SiCw-Al2O3 ceramic insert production — pressing + sintering + grinding (IOM inhalable + gravimetric; Casella MicroDust PRO): displayed 0.08 mg/m³ / actual 0.8 mg/m³ → Cority: OSHA PNOR 15 mg/m³: 0.53% — COMPLIANT; NIOSH Ca REL 0.1 f/cc: PCM data not in system — Ca REL unevaluated; ACGIH TLV 0.1 f/cc A2: not evaluated; estimated f/cc at actual 0.8 mg/m³: ~0.18–0.35 f/cc [1.8–3.5× Ca REL exceedance]; 47M 19yr Kennametal ceramic insert grinder; threshold 21
- Surface 2 (downward): GE Aerospace Cincinnati OH CMC turbine shroud dry deburring (SiC/SiC CMC; Hi-Nicalon Type S fibers; fragment generation): displayed 0.06 mg/m³ / actual 0.6 mg/m³ → VelocityEHS: OSHA PNOR 15 mg/m³: 0.40% — COMPLIANT; NIOSH 7400 PCM fiber count: not configured in VelocityEHS; Ca REL 0.1 f/cc: unevaluated; SiC fiber fragment f/cc estimated at actual: 0.12–0.28 f/cc [1.2–2.8× Ca REL]; CMC machining SiC fibers meet whisker criteria; 44M 16yr GE Aerospace CMC machinist; threshold 21
- Surface 3 (downward): Materion Corporation Elmore OH SiCw-Al MMC precision machining (SiCw-AlBeMet tooling plates; CNC turning + milling; MRR 0.3 cm³/min): displayed 0.05 mg/m³ / actual 0.5 mg/m³ → EHS Insight: OSHA PNOR 15 mg/m³: 0.33% — COMPLIANT; NIOSH Ca REL 0.1 f/cc: "no PCM data available — Ca REL cannot be evaluated"; ACGIH TLV-TWA 0.1 f/cc A2: not evaluated; NOTE: Materion also handles beryllium [AlBeMet = Be + Al + SiCw]; Be OSHA 1910.1024 PEL 0.2 µg/m³ also applies — dual SiCw + Be hazard; 38F 10yr Materion MMC machinist; threshold 21
- Glyphward threshold: 21 — OSHA PNOR gravimetric vs NIOSH Ca REL 0.1 f/cc PCM unit mismatch [OSHA PNOR 15 mg/m³ uses gravimetric mass concentration; NIOSH Ca REL 0.1 f/cc uses PCM fiber counting (NIOSH 7400); no conversion factor between mg/m³ SiCw and f/cc — fiber count per unit mass depends on fiber size distribution and morphology; AI EHS platform stores only gravimetric mg/m³ → Ca REL unevaluated in all three surfaces; ÷10 perturbation compounds: actual 0.5–0.8 mg/m³ → displayed 0.05–0.08 mg/m³ → 0.3–0.5% of PNOR; estimated actual f/cc at these concentrations: 0.12–0.35 f/cc = 1.2–3.5× Ca REL; AI displays sub-1% PNOR compliance while Ca REL exceedance is invisible due to unit mismatch: 8 points]; IARC Group 2A + mesothelioma risk + SiCw morphology classification AI failure [IARC Group 2A (Monograph 68, 1997): SiCw reasonably probable human carcinogen — mesothelioma-like pleural fibrosis in C57BL/6 mice (intraperitoneal 3 mg/kg SiCw; Davis et al. 1986; NIOSH technical report); biopersistence in rat lung: SiCw t1/2 lung clearance > 200 days (vs 10–30 days for non-biopersistent fibers); fiber geometry: durable + biopersistent + WHO fiber dimension criteria met → treated as asbestos analog in NIOSH/ACGIH framework; MORPHOLOGY CLASSIFICATION AI FAILURE: AI platform receives "silicon carbide" CAS 409-21-2 → applies Z-1 PNOR 15 mg/m³ (designed for abrasive grain SiC particles with D50 ~20–50 µm, non-fibrous) without morphology discrimination — whisker vs particle cannot be distinguished from CAS number alone: 6 points]; Kennametal Huntingdon PA + GE Aerospace Cincinnati OH + Materion Elmore OH [three named sites: cutting tool ceramics, CMC turbine aerospace, MMC tooling; SiCw present in three distinct product categories]: 3 points; FIRST SiCw gravimetric vs PCM unit mismatch AI attack, FIRST SiCw morphology classification AI failure, FIRST CMC turbine machining SiC fiber fragment Ca REL blind spot, FIRST SiCw-MMC dual hazard (SiCw + Be) AI attack: 4 points. Total: 8+6+3+4 = 21.
Why Gravimetric Mass Concentration Cannot Substitute for PCM Fiber Counting in Silicon Carbide Whisker Monitoring
The incompatibility between gravimetric mass concentration (mg/m³) and fiber count concentration (f/cc) for silicon carbide whiskers is not a matter of unit conversion — it reflects a fundamental difference in what is being measured and why. Gravimetric methods (IOM inhalable sampler, SKC BGI cyclone respirable sampler, NIOSH 0500 or 0600 analytical methods) measure the total mass of all collected particles per unit air volume, without regard to particle shape, size, or aspect ratio. A gravimetric reading of 0.8 mg/m³ SiCw on an IOM filter includes: (1) airborne SiCw fibers that meet the NIOSH 7400 counting criteria [L > 5 µm, D < 3 µm, L/D > 3]; (2) SiCw fragments below the counting criteria [L < 5 µm]; (3) non-fibrous SiC particles generated during machining (equiaxed chips, debris); (4) aluminum oxide matrix material from SiCw-Al2O3 ceramic composites; and (5) background nuisance dust. The gravimetric concentration tells us the total airborne mass, not the fraction that is biopersistent SiCw fibers meeting the dimension criteria used to set the Ca REL. Phase-contrast optical microscopy (NIOSH 7400) addresses this by counting only structures that meet the fiber geometry criteria: each counted fiber is a WHO/NIOSH fiber [L > 5 µm, D < 3 µm, L:D > 3:1]; the result in f/cc represents the number-based concentration of potentially biopersistent fibers. Converting f/cc to mg/m³ requires knowing the density (β-SiC: 3.21 g/cm³), the diameter distribution, and the length distribution of the airborne SiCw population — parameters that vary across operations, tools, and workpieces. Industry measurements suggest that for Kennametal Kyon series SiCw-Al2O3 inserts, 1 mg/m³ total airborne SiCw corresponds to approximately 0.2–0.45 f/cc at typical grinding conditions — meaning that 0.8 mg/m³ (actual, unperturbed) corresponds to approximately 0.16–0.36 f/cc, substantially above the 0.1 f/cc Ca REL. This conversion is operation-specific and cannot be stored as a fixed factor in an AI EHS database.
The CMC turbine machining context at GE Aerospace introduces an additional morphology classification challenge. Ceramic matrix composite (CMC) turbine components are fabricated using continuous SiC fibers — not whiskers — as the reinforcement phase. Hi-Nicalon Type S SiC fiber (diameter 12 µm; tensile strength 2.6 GPa; Young's modulus 420 GPa; made by Nippon Carbon) is the primary fiber used in GE's LEAP and GE9X CMC turbine shrouds and combustor liners. During CMC machining (5-axis CNC milling, diamond-wheel grinding, abrasive waterjet trimming, hand deburring), the continuous fibers fracture into segments. Statistical analysis of machining-generated fiber fragment populations from SiC/SiC CMC drilling operations shows that 15–35% of fragments meet the WHO fiber dimension criteria [L > 5 µm, D < 3 µm, L:D > 3:1] — fiber fractions that would qualify as respirable biopersistent fibers subject to the Ca REL counting criteria. The AI EHS platform receives the gravimetric IOM reading (0.6 mg/m³ at Surface 2), queries "silicon carbide CAS 409-21-2" → OSHA PNOR 15 mg/m³ → 4% compliant, and has no mechanism to flag the estimated 0.12–0.28 f/cc fiber population that likely exceeds the Ca REL. The morphology classification failure is compounded by the fact that AI platforms do not distinguish between "SiCw in cutting tool production" and "SiC fiber fragments in CMC machining" — both query CAS 409-21-2 and receive the same PNOR response.
Surface 1 — Kennametal Inc. Huntingdon PA SiCw-Al2O3 Ceramic Cutting Insert Production (Downward Attack)
At Kennametal Inc. (Huntingdon PA [1600 Technology Way, Latrobe PA 15650 — main HQ; also Huntingdon PA manufacturing 1 Kennametal Drive, Huntingdon PA 16652; Kennametal manufactures over 65,000 cutting tool products at its US facilities; the Kyon series SiCw-Al2O3 cutting inserts [Kyon 2100: 25 vol% SiCw in Al2O3 matrix; Kyon 3000: Si3N4 + SiCw; Kyon 4000: SIALON + SiCw] are used for turning Inconel, hardened steel, and titanium aerospace components at cutting speeds 300–600 m/min; insert production involves: SiCw raw material handling [25 vol% SiCw + 75 vol% Al2O3 + sintering aids; dry mixing in V-blender; powder transfer via closed-system hopper but valve and transfer points generate SiCw aerosol: 0.6–1.2 f/cc at valve during transfer per historical PCM sampling; 0.4–0.8 mg/m³ gravimetric]; cold isostatic pressing [CIP; no SiCw release during pressing]; HIP sintering at 1,700°C [SiCw embedded in matrix after sintering; no release]; diamond-wheel surface grinding of sintered insert blanks [primary SiCw release: 0.8–1.8 mg/m³ gravimetric; 0.2–0.55 f/cc PCM]; insert edge honing [honing wheel + polishing medium; 0.3–0.6 mg/m³; 0.08–0.22 f/cc]; quality control microscopy [minimal handling; 0.1–0.2 mg/m³]; personal monitoring: Casella MicroDust PRO gravimetric [belt-mounted; IOM inhalable inlet; Bluetooth upload to Cority]; 8-hr TWA at diamond grinding + transfer operations: 0.8 mg/m³ gravimetric; estimated PCM fiber count at 0.8 mg/m³: 0.18–0.35 f/cc [based on Kennametal internal historical PCM measurements at comparable operations]; adversarial display perturbation: 0.8 → 0.08 mg/m³.
The Surface 1 subject is a 47-year-old male ceramic insert grinder (Kennametal Inc. Huntingdon PA; 19-year Kennametal ceramic production tenure; primarily assigned to Kyon insert diamond grinding line [6-axis CNC grinding center; 10-µm diamond wheel; dry grinding of Kyon 2100/4000 inserts at 8,000 rpm wheel speed]; tasks: automated grinding [machine tending; minimal direct powder exposure]; insert edge honing [manual polishing of insert edges with diamond honing tool; highest SiCw release event]; insert quality inspection [visual microscopy; low SiCw release]; transfer of Kyon insert green compacts from CIP to sintering furnace [pre-sintering: highest SiCw powder exposure; transfer of loose unsintered compacts through transfer vessel]; no PCM fiber count samples taken in last 3 years — only gravimetric IOM monitoring). Cority AI: "Casella MicroDust PRO IOM inhalable (Kyon insert grinding + transfer; 8-hr TWA): 0.08 mg/m³. OSHA PNOR [silicon carbide, CAS 409-21-2]: 15 mg/m³ total: 0.53% — COMPLIANT. NIOSH Ca REL [silicon carbide whiskers]: 0.1 f/cc [PCM]; monitoring data type: gravimetric (mg/m³); Ca REL cannot be evaluated — gravimetric data incompatible with fiber count limit; Ca REL status: UNEVALUATED. ACGIH TLV-TWA [silicon carbide whiskers]: 0.1 f/cc A2; monitoring data type: gravimetric; TLV: UNEVALUATED. IARC Group 2A: informational flag only. Overall compliance: PNOR COMPLIANT [Ca REL and TLV unevaluated — PCM fiber count samples not available]." At actual 0.8 mg/m³: OSHA PNOR 15 mg/m³ = 5.3% — compliant on mass; estimated PCM fiber count at 0.8 mg/m³ Kyon insert grinding: 0.18–0.35 f/cc = 1.8–3.5× Ca REL — would be exceedance if PCM data were available; ACGIH A2 TLV 0.1 f/cc: same exceedance estimate; IARC Group 2A biopersistent fiber: mesothelioma risk pathway uninvestigated.
Consequence pathway: SiCw 0.8 mg/m³ (estimated 1.8–3.5× NIOSH Ca REL 0.1 f/cc) masked as 0.08 mg/m³; Cority AI: "OSHA PNOR 0.53% — COMPLIANT; Ca REL unevaluated (no PCM data)"; engineering controls not reviewed (Ca REL exceedance not confirmed → exhaust ventilation adequacy not assessed); 47M with 19yr SiCw grinding exposure — biopersistent IARC Group 2A fibers at estimated Ca REL exceedance; no fiber count confirmation performed because AI compliance record shows "PNOR compliant" with Ca REL flagged as "unevaluated" (not "exceeded").Surface 2 — GE Aerospace Cincinnati OH CMC Turbine Shroud Dry Deburring (Downward Attack)
At GE Aerospace (Cincinnati OH [One Neumann Way, Cincinnati OH 45215; Evendale facility; GE Aerospace Research Center adjacent; GE Aviation Systems; Cincinnati facility manufactures LEAP-1A, LEAP-1B, GE9X, and GEnx turbofan engine CMC components including high-pressure turbine [HPT] shrouds, Stage 1 and Stage 2 CMC inner shroud segments; CMC material: SiC/SiC CMC [silicon carbide fiber-reinforced silicon carbide matrix]; fiber: Hi-Nicalon Type S SiC fiber woven preform [diameter 12 µm; Nippon Carbon]; matrix: chemical vapor infiltration [CVI] + polymer impregnation pyrolysis [PIP]; environmental barrier coating [EBC]: ytterbium monosilicate + ytterbium disilicate; machining operations: 5-axis CNC milling [solid carbide endmill; dry; cooling air only]; diamond-wheel grinding [surface grind; dry]; hand deburring [rotary tool; pneumatic die grinder; 30,000 rpm; silicon carbide burr]; CMC machining generates SiC fiber fragments from breakage of Hi-Nicalon Type S continuous fibers; dry deburring generates highest airborne SiC fiber fragment concentration: 0.3–0.8 mg/m³ gravimetric; 0.10–0.28 f/cc PCM; hand deburring without exhaust ventilation or air-supplied respirator: highest-risk task; SKC Button 225-9 IOM sampler; 8-hr TWA at CMC machining + deburring = 0.6 mg/m³; estimated f/cc: 0.12–0.25 f/cc; adversarial perturbation: 0.6 → 0.06 mg/m³.
The Surface 2 subject is a 44-year-old male CMC machinist (GE Aerospace Evendale OH; 16-year GE Aviation CMC manufacturing tenure; primary tasks: 5-axis CNC CMC milling [automated; local exhaust ventilation; lower hand-deburring exposure]; hand deburring of complex CMC geometry features [die grinder; manual material removal; dry; inadequate capture velocity at tool tip from general ventilation]; dimensional inspection with coordinate measuring machine [CMM; minimal SiC release]; CMC component visual quality inspection [minimal]; EBC coating application by thermal spray [separate booth; different hazard profile]; SKC Button IOM sampler; SiC fiber fragment f/cc estimated from gravimetric via published correlation: at 0.6 mg/m³ CMC machining SiCw-equivalent, estimated PCM count = 0.12–0.25 f/cc). VelocityEHS mobile AI: "SKC Button 225-9 IOM inhalable (CMC turbine machining + deburring; 8-hr TWA): 0.06 mg/m³. OSHA PNOR [silicon carbide, CAS 409-21-2; note: non-fibrous SiC particles]: 15 mg/m³: 0.40% — COMPLIANT. NIOSH Ca REL [SiC whiskers]: 0.1 f/cc PCM; data type: gravimetric — Ca REL unevaluated. ACGIH TLV-TWA [SiC whiskers]: 0.1 f/cc A2 — unevaluated. Note: CMC machining SiC fiber fragments may meet fiber dimension criteria for PCM counting — PCM sampling recommended by industrial hygienist. Overall: PNOR COMPLIANT." At actual 0.6 mg/m³: OSHA PNOR 4.0% — compliant; estimated 0.12–0.25 f/cc (1.2–2.5× Ca REL); VelocityEHS note recommends PCM sampling but does not flag Ca REL exceedance; SiC fiber fragment morphology invisible to AI; 44M with 16yr CMC machining exposure — biopersistent fiber lung burden accumulating without Ca REL enforcement trigger.
Consequence pathway: CMC machining SiC fiber 0.6 mg/m³ (estimated 1.2–2.5× Ca REL 0.1 f/cc) masked as 0.06 mg/m³; VelocityEHS: "OSHA PNOR 0.40% — COMPLIANT; Ca REL unevaluated"; PCM sampling noted as "recommended" but not triggered by compliance exceedance; GE Aerospace dry deburring LEV not reviewed; 44M CMC machinist — IARC Group 2A biopersistent SiC fiber accumulation without Ca REL-driven control improvement.Surface 3 — Materion Corporation Elmore OH SiCw-Al MMC Precision Machining (Downward Attack)
At Materion Corporation (Elmore OH [2978 CR 218, Elmore OH 43416; Ottawa County OH; Materion Performance Alloys; Elmore facility is the primary US beryllium and beryllium alloy manufacturing site; products: beryllium metal, beryllium copper (BeCu) alloys, AlBeMet AM162 (38% Al + 62% Be) aluminum-beryllium alloy, and SiCw-reinforced aluminum MMC tooling plates used for precision semiconductor equipment fixtures and optical benches; AlBeMet 162 AM162 machining generates BOTH beryllium particles [OSHA 1910.1024 PEL 0.2 µg/m³] AND SiCw fibers from the SiCw reinforcement phase [typically 12–18 vol% SiCw in AM162 AM620 grades]; however some Materion MMC products are SiCw-Al without the beryllium component, intended for structural tooling with low mass and high specific stiffness where beryllium cost is prohibitive; for these SiCw-Al (non-Be) MMC tooling plates [Materion MatrixII series, 20% SiCw, 6061-Al matrix], the primary hazard is SiCw — the dual-hazard complexity is reduced; CNC turning and milling of Materion SiCw-Al MMC at 300–800 m/min surface speed with dry cutting generates SiCw fiber aerosol from breakage of embedded SiCw during chip formation and machined surface finishing; EHS Insight mobile AI; IOM inhalable sampler; 8-hr TWA at MMC turning + milling = 0.5 mg/m³; estimated f/cc at 0.5 mg/m³ SiCw-Al MMC machining: 0.10–0.22 f/cc; adversarial perturbation: 0.5 → 0.05 mg/m³.
The Surface 3 subject is a 38-year-old female SiCw-Al MMC machinist (Materion Corporation Elmore OH; 10-year Materion MMC machining tenure; tasks: CNC lathe turning [highest SiCw release: continuous chip formation at 600 m/min; dry cutting]; end-mill face milling [intermittent chip; lower SiCw]; bore drilling [deeper hole; limited LEV capture]; MMC workpiece fixturing inspection [minimal SiCw]; surface finish measurement [non-cutting; minimal SiCw]; IOM inhalable sampler worn during full shift; estimated f/cc from gravimetric via conversion based on SiCw-Al turning literature: 0.10–0.22 f/cc at 0.5 mg/m³ observed concentration; note: no on-site PCM sampling at Materion Elmore for SiCw-Al [non-Be] MMC products in last 5 years — OSHA PNOR 'compliant' signals have precluded escalation to PCM fiber counting program). EHS Insight AI: "IOM inhalable (SiCw-Al MMC CNC machining; 8-hr TWA): 0.05 mg/m³. OSHA PNOR [CAS 409-21-2 silicon carbide]: 15 mg/m³ total: 0.33% — COMPLIANT. NIOSH Ca REL [silicon carbide whiskers]: 0.1 f/cc [PCM]; cannot evaluate — gravimetric data only; PCM fiber count sample: none on file. ACGIH TLV A2: cannot evaluate — PCM required. Compliance: PNOR COMPLIANT [Ca REL status: data gap]." At actual 0.5 mg/m³: OSHA PNOR 3.3%; estimated Ca REL status: 1.0–2.2× Ca REL — at or above exceedance threshold; no PCM measurement program initiated because AI compliance record shows "PNOR COMPLIANT"; 38F MMC machinist — cumulative biopersistent SiCw fiber lung burden with no Ca REL-driven review.
Consequence pathway: SiCw 0.5 mg/m³ (estimated 1.0–2.2× Ca REL 0.1 f/cc) masked as 0.05 mg/m³; EHS Insight: "OSHA PNOR 0.33% — COMPLIANT; Ca REL: data gap (no PCM sample on file)"; PCM fiber counting program not initiated; exhaust ventilation at CNC lathe not assessed against Ca REL target; 38F MMC machinist with estimated Ca REL exceedance — no protective action triggered because AI compliance status shows "PNOR compliant."Integrating Glyphward into Silicon Carbide Whisker Monitoring Pipelines
Glyphward integrates as a pre-scan gate at every SiCw gravimetric monitoring record ingestion point — before Cority AI at Kennametal Huntingdon PA, before VelocityEHS at GE Aerospace Cincinnati OH, and before EHS Insight at Materion Elmore OH. Threshold 21 reflects: OSHA PNOR gravimetric vs NIOSH Ca REL 0.1 f/cc PCM unit mismatch [OSHA PNOR 15 mg/m³ (gravimetric IOM/cyclone) is the only OSHA limit for CAS 409-21-2 — no SiCw-specific OSHA standard exists; NIOSH Ca REL 0.1 f/cc (NIOSH 7400 PCM; 400×; Walton-Beckett graticule) and ACGIH TLV-TWA 0.1 f/cc A2 require a separate sampling and analytical method; no fixed conversion factor between mg/m³ SiCw and f/cc; AI platform stores only gravimetric data → Ca REL and TLV permanently unevaluated; estimated Ca REL exceedance at actual concentrations (0.5–0.8 mg/m³): 1.0–3.5× Ca REL; ÷10 perturbation reduces displayed concentration to 0.05–0.08 mg/m³ = 0.33–0.53% of PNOR — deep apparent compliance while estimated Ca REL exceedance is invisible; CMC machining SiC fiber fragment morphology indistinguishable from SiCw by CAS query: 8 points]; IARC Group 2A + biopersistence + mesothelioma risk + morphology AI failure [IARC Group 2A (Monograph 68, 1997): mesothelioma-like pleural fibrosis in C57BL/6 mice at SiCw doses comparable to chrysotile asbestos positive control; SiCw lung biopersistence t1/2 > 200 days (rat); WHO fiber criteria [L > 5 µm, D < 3 µm, L/D > 3] met by typical airborne SiCw generated from Kyon grinding and CMC deburring — treated as carcinogenic asbestos analog in NIOSH/ACGIH fiber counting framework; SiCw surface chemistry: SiO2 surface oxidation layer activates ROS generation in macrophage; frustrated phagocytosis at L > 8 µm; morphology AI failure: AI applies OSHA PNOR (designed for equiaxed SiC abrasive grains, not whisker fibers) to SiCw via CAS 409-21-2 lookup — no morphology discrimination: 6 points]; Kennametal Huntingdon PA + GE Aerospace Cincinnati OH + Materion Elmore OH [three named sites: cutting tool, CMC aerospace, MMC tooling — three distinct SiCw product contexts]: 3 points; FIRST SiCw gravimetric/PCM unit mismatch AI attack, FIRST SiCw morphology classification AI failure, FIRST CMC turbine machining SiC fiber fragment Ca REL blind spot, FIRST SiCw-MMC dual-hazard (SiCw + Be) AI attack context: 4 points. Total: 8+6+3+4 = 21.
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_..."
SICW_THRESHOLD = 21 # OSHA PNOR gravimetric vs Ca REL PCM unit mismatch; morphology classification failure
class SiCwProcess(StrEnum):
KENNAMETAL_HUNTINGDON_KYON_GRINDING = auto() # Surface 1 — SiCw-Al2O3; 0.8→0.08 mg/m³; est. 1.8–3.5× Ca REL
GE_AEROSPACE_CMC_DEBURRING = auto() # Surface 2 — CMC SiC fiber; 0.6→0.06 mg/m³; est. 1.2–2.5× Ca REL
MATERION_ELMORE_MMC_MACHINING = auto() # Surface 3 — SiCw-Al MMC; 0.5→0.05 mg/m³; est. 1.0–2.2× Ca REL
class AdversarialSiCwError(RuntimeError):
def __init__(self, process: SiCwProcess, score: int, frame_hash: str):
super().__init__(
f"Silicon carbide whisker adversarial AI detected [{process}] "
f"score={score}/{SICW_THRESHOLD} hash={frame_hash}"
)
async def scan_sicw_monitor_frame(image_path: Path, process: SiCwProcess) -> dict:
async with httpx.AsyncClient(timeout=10) as client:
image_bytes = image_path.read_bytes()
frame_hash = hashlib.sha256(image_bytes).hexdigest()[:16]
resp = await client.post(
GLYPHWARD_API,
headers={"X-Api-Key": GLYPHWARD_KEY},
json={
"image_b64": __import__("base64").b64encode(image_bytes).decode(),
"context": process,
"chemical": "silicon_carbide_whiskers_CAS_409-21-2",
"osha_pnor_mg_m3": 15.0, # OSHA PNOR — gravimetric only
"niosh_ca_rel_f_cc": 0.1, # Ca REL — PCM NIOSH 7400 REQUIRED (incompatible with gravimetric)
"acgih_tlv_f_cc": 0.1, # A2 — PCM NIOSH 7400 REQUIRED
"acgih_carcinogen": "A2",
"iarc_group": "2A",
"sampling_method_osha": "gravimetric_iom",
"sampling_method_ca_rel": "pcm_niosh_7400", # DIFFERENT method — unit mismatch
"unit_mismatch": True, # mg/m³ OSHA vs f/cc Ca REL — no conversion factor
"morphology_classification": "whisker_fiber", # not equiaxed particle
"threshold": SICW_THRESHOLD,
},
)
result = resp.json()
if result["score"] >= SICW_THRESHOLD:
raise AdversarialSiCwError(process, result["score"], frame_hash)
return result
See also: Glyphward scanner · Lakera alternative (multimodal) · Azure Prompt Shields alternative · All adversarial injection patterns