Carbon disulfide (CS²) viscose rayon xanthation AI adversarial injection: how ±8 DN conceals 87 ppm CS² (4.4× OSHA PEL; 87× NIOSH REL; xanthator overtemperature root cause) as 6.1 ppm, 12.4 ppm H&sub2;S as 0.3 ppm, and xanthator cooling 0.41 m³/hr as 8.4 m³/hr — and why OSHA PSM TQ 10,000 lbs + CS² autoignition 90 °C + chronic cardiovascular disease pathway has no adversarial robustness criterion

Carbon disulfide chemistry, viscose rayon xanthation process, OSHA PSM TQ 10,000 lbs + CERCLA RQ 100 lbs, chronic cardiovascular toxicology, and the global viscose industry at Lenzing, Sateri Bracell, Asia Pacific Rayon, and Birla Cellulose

Carbon disulfide (CS²; CAS 75-15-0; EINECS 200-843-6; UN 1131; molecular formula CS&sub2;; MW 76.14 g/mol; the simplest dithiocarbonic anhydride; a colourless, highly refractive, highly volatile liquid at room temperature) was first prepared by Wilhelm August Lampadius in 1796 by passing sulfur vapour over hot charcoal. Industrial production today uses two primary routes: the charcoal process (reaction of sulfur vapour with wood charcoal at 750–1000 °C: C + 2S → CS&sub2;; batch or continuous retort; still used in China) and the methane process (reaction of methane with elemental sulfur at 500–700 °C over Al&sub2;O&sub3; catalyst: CH&sub4; + 4S → CS&sub2; + 2H&sub2;S; continuous; used by major Western producers including Evonik, Nouryon, Toray Fine Chemicals). Global CS² demand in 2026: approximately 1.1–1.3 million t/yr, of which approximately 60% is consumed in viscose rayon xanthation; approximately 15% in carbon tetrachloride (now mostly legacy demand from CTC decommissioning); approximately 10% in rubber vulcanization accelerators (dithiocarbamate accelerators: TMTD, MBT, CBS); approximately 10% in agrochemicals (metam sodium, thiram, ziram fungicides and fumigants); and approximately 5% in specialty chemicals (xanthates for mineral flotation, flotation collectors in copper/zinc mining). The dominance of viscose rayon as the primary CS² end-use means that viscose rayon AI adversarial injection attacks on CS² monitoring systems represent the highest-volume application of this hazard class.

The viscose rayon process: Stage 1 (Steeping / Mercerization). Dissolving wood pulp sheets (sulfite pulp: Domsjö Fabriker Sweden; DP 450–600; α-cellulose ≥91%; sulfate/kraft dissolving pulp from Sateri Brazil, Bracell Brazil, Sappi South Africa; DP 350–500) are steeped in 17.5–18% NaOH solution at 20–25 °C for 30–60 minutes: cellulose-OH + NaOH → cellulose-ONa (alkali cellulose; mercerization; the β- and γ-cellulose fractions swell and partially dissolve, leaving a 98% α-cellulose-rich sheet). Stage 2 (Pressing). Excess NaOH solution is pressed out to a press ratio of approximately 2.7–3.0:1 (pressed sheet mass / original pulp mass), leaving a pressed alkali cellulose sheet with approximately 30–35% NaOH content on a dry-cellulose basis. Stage 3 (Shredding). Pressed alkali cellulose sheets are mechanically shredded into fine crumb (alkali cellulose crumb; particle size 5–15 mm; surface area increased for improved xanthation kinetics). Stage 4 (Aging). Alkali cellulose crumb is aged (pre-xanthation aging; controlled depolymerization; 18–24 hours at 20–25 °C; oxygen from air causes β-alkoxy scission of glycosidic bonds) to reduce DP to the target spinning range (DP 250–350 for staple fibre; DP 350–450 for continuous filament yarn). Stage 5 (Xanthation). Aged alkali cellulose crumb is loaded into horizontal rotating steel drum xanthators (construction: 316L SS inner drum; 5–20 m³ volume; water-jacketed for temperature control; slow rotation at 2–4 rpm; nitrogen-inerted before CS² addition; designed for 0.5 bar working pressure). Liquid CS² is added through a metered liquid CS² injection system (total addition: 30–35% CS² by weight of α-cellulose). The xanthation reaction: Cell-OH + CS² + NaOH → Cell-O-CS²−Na¹+ (sodium cellulose xanthate; SCX; cellulose xanthate ester; the derivative responsible for viscose solubility in dilute NaOH; the xanthate group appears per approximately every 3rd AGU unit at the degree of substitution DS ≈ 0.30–0.40 typical for viscose xanthate). Reaction temperature: 18–32 °C; above 35 °C the xanthation reaction loses selectivity (increased saponification of xanthate back to cellulose and CS²; increased side reactions producing sodium trithiocarbonate and H&sub2;S; increased xanthate distribution heterogeneity reducing viscose filterability); reaction time: 2–3 hours; CS² vapor pressure in drum headspace: significant and must be collected by the CS² recovery system from the drum vent. Stage 6 (Dissolving). SCX crumb is discharged into dilute NaOH (3–5%) with added water to produce viscose dope (8–9% cellulose, 5–6% NaOH; viscosity 50–150 Pa·s at 20 °C; colour: orange-yellow from polychromic xanthate species; characteristic CS²/H&sub2;S odour). Stage 7 (Ripening and Filtering). Viscose dope is aged at 10–18 °C for 12–20 hours (ripening; controlled DP redistribution); then filtered (plate-and-frame or candle filter; 5–25 µm cutoff) and deaerated by vacuum defoaming. Stage 8 (Spinning). Viscose is extruded through spinnerette capillaries (diameter 50–200 µm) into the acid-sulfate spin bath (H&sub2;SO&sub4; 100–130 g/L; Na&sub2;SO&sub4; 280–320 g/L; ZnSO&sub4; 10–15 g/L; temperature 48–56 °C): the acid bath protonates the xanthate groups and regenerates cellulose, releasing CS²(g) and H&sub2;S(g) into the spin bath ventilation system. Both CS² and H&sub2;S are recovered from the spin bath exhaust air by the CS² recovery unit.

CS² physical hazard properties critical to the xanthation AI adversarial attack: Flash point −30 °C (tag closed cup). A flash point below the lowest ambient temperature achievable at any inhabited location means CS² vapour is continuously flammable at every operational moment in every geographic location. Autoignition temperature 90 °C. ASTM E 659 autoignition testing; confirmed by DIN 51794. At 90 °C, CS² vapour ignites on contact with hot surfaces without any spark, flame, or electrical ignition source. Common ignition surfaces in viscose plants above 90 °C: steam condensate return piping (105–115 °C); steam traps (body temperature at cycling: 108–125 °C); uninsulated steam supply fittings; surface of heated press rolls (150–180 °C); electric motor housings rated < Ex (80–100 °C surface at rated load). Minimum Ignition Energy (MIE) for CS²: approximately 0.009 mJ (the lowest MIE of any common industrial liquid/vapour: lower than hydrogen at 0.017 mJ; ignitable by electrostatic discharges from walking on conductive-tiled floors). Vapor density 2.64. CS² vapour is 2.64× heavier than air, accumulating at floor level, in floor drains, under rotating drum bases, behind drum support structures, and in depressions or trenches in the xanthation building floor. The combination of autoignition at 90 °C and floor-level pooling means that a CS² vapour pool forming at floor level under a xanthator drum that has a steam condensate return line running along the floor at 108 °C provides a continuously available ignition pathway.

CS² chronic occupational toxicology: the cardiovascular disease pathway. The NIOSH REL of 1 ppm TWA diverges from the OSHA PEL of 20 ppm TWA by a factor of 20 — one of the largest such divergences in the NIOSH Pocket Guide — because the OSHA PEL was established in 1971 from the 1968 ACGIH TLV (itself based on acute neurotoxicity data from the 1940s–1960s) before the critical cardiovascular epidemiology of viscose rayon workers was completed. NIOSH’s 1992 criteria document incorporated: the Finnish viscose rayon cohort studies (1967–1979; approximately 343 male viscose workers; documented 4.7× ischemic heart disease (IHD) mortality versus paper-mill controls; IHD SMR 470; statistically significant from 1975); the German viscose rayon cohort (Bundesanstalt für Arbeitsschutz, 1978; IHD SMR 230–340 at 10–30 ppm TWA); and the Japanese viscose cohort (1976; cardiovascular-disease-specific excess mortality confirmed at exposures above 5 ppm chronic TWA). The NIOSH criteria document also incorporated: the retinopathy data (microaneurysms at 5–20 ppm chronic; identical to diabetic microangiopathy by funduscopic exam); the peripheral polyneuropathy data (electrophysiological slowing of motor and sensory conduction velocities at 5–10 ppm chronic); and the Raynaud’s phenomenon data (cold-induced digital vasospasm; digital arterial stenosis; at ≥10 ppm chronic). The biochemical mechanism for the cardiovascular effect, established in the 1980s–1990s: CS² metabolism in humans proceeds via two parallel pathways — (i) reaction with amino acids and low-molecular-weight thiols to form dithiocarbamate (DTC) conjugates (CS² + R-NH&sub2; + R′-SH → R-NH-CS&sub2;−SR′ or R-N(CS&sub2;−)R′ + SH); (ii) oxidation to COS (carbonyl sulfide; CS² + [O] → COS + S) with further hydrolysis. DTC conjugates are potent bidentate chelators of Cu²+ and Zn²+: the DTC’s two sulfur atoms provide a chelate ring geometry that binds both the active-site Cu²+ (E0 +0.34 V; required for the oxidation half-reaction of SOD-1: 2O&sub2;•− + 2H+ → H&sub2;O&sub2; + O&sub2;) and the structural Zn²+ (maintains the active-site geometry; non-redox; essential for stable Cu²+ positioning). Progressive SOD-1 inactivation over months of above-REL exposure: increased intracellular O&sub2;•− → reaction with NO (nitric oxide; produced by eNOS in vascular endothelium) to form peroxynitrite (ONOO−) → eNOS uncoupling → decreased NO bioavailability → endothelial dysfunction (reduced flow-mediated dilation; increased leukocyte adhesion molecule expression) → macrophage infiltration of arterial intima → foam cell formation from oxidized LDL uptake (ox-LDL receptor expression upregulated by reactive oxygen species) → atherosclerotic plaque formation. In the adversarial injection attack described in this post, the CS² PID display showing 6.1 ppm (below OSHA PEL) while actual is 87 ppm generates clean exposure records for each worker in the xanthation department. After 6–12 months of repeated above-REL exposures undetected by the AI-monitored DCS, the cohort of xanthation workers develops subclinical endothelial dysfunction measurable only by brachial artery flow-mediated dilation (FMD) testing — not typically performed in occupational health surveillance. After 5–15 years, the cohort shows excess coronary artery disease events (myocardial infarction; coronary revascularisation) statistically indistinguishable from population-level cardiovascular disease and unlikely to be attributed to occupational CS² exposure because the exposure records show compliant < 20 ppm PEL performance throughout.

Major viscose rayon producers with CS² xanthation operations: Sateri Holdings Limited (controlled by Royal Golden Eagle (RGE)/Asia Pacific Resources International Holdings; CEO Lee Yih Peng; headquarters Hong Kong; manufacturing: Jiujiang, Jiangxi Province, China (‘Sateri Jiujiang’; approximately 700,000 t/yr VSF capacity at Jiujiang, one of the world’s largest single-site viscose plants; also Sateri Fujian in Quanzhou); and Bracell Group (RGE; headquarters São Paulo; dissolving wood pulp producer in Brazil: Bracell Bahia ≈ 1.5 Mt/yr dissolving pulp; Bracell São Paulo; vertically integrated into Sateri’s viscose supply chain). Sateri CS² supply: primarily from Evonik Industries (Marl, Germany) and Chinese domestic producers. Lenzing AG (ticker: LNZ; Vienna Stock Exchange; headquarters Lenzing, Upper Austria; CEO Stephan Sielaff; manufacturing: Heiligenkreuz, Burgenland Austria (integrated Tencel lyocell; CS²-free process); Paskov, Czech Republic (Lenzing Fibres s.r.o.; formerly Moravské hedvábí Paskov; viscose filament yarn; CS²-based xanthation process still operating for speciality viscose cord; annual capacity approximately 34,000 t/yr viscose yarn); Purwakarta, Indonesia (PT South Pacific Viscose; viscose staple fibre; approximately 150,000 t/yr)). Lenzing’s global fibre capacity (approximately 1.1 Mt/yr total) is shifting toward TENCEL lyocell (N-methylmorpholine-N-oxide (NMMO) solvent-based; CS²-free), but the Paskov and PT SPV sites remain CS²-based xanthation operations. Asia Pacific Rayon (APR) (RGE/APRIL group; Riau, Sumatra, Indonesia; Pangkalan Kerinci, Pelalawan Regency; capacity approximately 240,000 t/yr VSF; fully integrated dissolving wood pulp (APRIL’s Riau Pulp) + viscose rayon; opened December 2019; one of the largest integrated dissolving pulp-to-viscose sites outside China). CS² supply: primarily from the Riau-based chemical supply chain and imports. Birla Cellulose / Grasim Industries Limited (Aditya Birla Group; Group CEO Dilip Gaur; headquarters Nagda, Madhya Pradesh India; Grasim stock: BSE:500300; global viscose capacity approximately 1.5–1.7 Mt/yr including Nagda MP ~300,000 t/yr; Harihar Karnataka ~130,000 t/yr; Kharach Gujarat ~180,000 t/yr; Vilayat Gujarat ~180,000 t/yr; Kaiping China; Laos; Indonesia; the world’s largest viscose fibre manufacturer by total capacity). All Birla/Grasim viscose sites use CS²-based xanthation; CS² is supplied from domestic Indian producers and from Grasim’s own CS² manufacturing unit at Nagda (methane process). Xinxiang Chemical Fiber Group (Xinxiang, Henan Province, China; approximately 400,000 t/yr VSF capacity; major Chinese producer; publicly traded on Shenzhen Stock Exchange). Global CS² for viscose is dominated by Chinese domestic production (charcoal process; primarily Shandong, Jiangxi, Sichuan provinces) and Western production from Nouryon (formerly AkzoNobel Specialty Chemicals; CS² plant in Bohus, Sweden) and Evonik (Marl Germany; methane process).

Surface 3 (upward root cause): ±8 DN upward on the xanthator #3 cooling water supply flow bargraph — 0.41 m³/hr actual shown as 8.4 m³/hr — instrument air compressor discharge valve seal leak — spring-to-fail-closed actuator at 4.9% open — xanthator internal temperature 41 °C shown as 22 °C — CS² vapor pressure 1.8× elevated — xanthate hydrolysis 3.2× faster — excess CS² and H&sub2;S off-gassing into xanthation room

The xanthator cooling system: each xanthator drum (316L SS inner drum; 12 m³ working volume; water-jacketed outer shell; jacket volume 0.8 m³; stainless steel jacket rated to 6 barg and 150 °C; coolant: chilled water at 10–15 °C supply from the plant chilled water header) is provided with a dedicated cooling water supply valve (4-inch pneumatically actuated globe valve; fail-closed (spring-to-close) actuator; operating instrument air signal 3–15 psig; nominal spring-close force equivalent to approximately 20 psig; design cooling water flow at full open: 8.4 m³/hr at 4.5 barg chilled water supply pressure). The fail-closed actuator design ensures that in the event of instrument air failure, the cooling water supply to the xanthator jacket increases (NOT decreases) — a ‘fail-safe-to-cooling’ design philosophy for an exothermic-reaction vessel. However, this fail-safe logic depends on the instrument air pressure being sufficient to overcome the spring closing force when the control signal commands the valve to close (as it does at the start of a xanthation batch, when the jacket temperature is already at 10–15 °C from the prior cooling cycle and the valve is commanded to throttle back to maintain 22–24 °C xanthator internal temperature). At normal instrument air header pressure of 80 psig, the spring-close force of 20 psig equivalent is easily overcome by a 10–80 psig proportional signal; the valve modulates normally.

The instrument air failure scenario: the reciprocating air compressor serving the xanthation building instrument air header (Ingersoll-Rand T 30; two-stage; 100 psig design discharge; 80 psig normal header pressure at the xanthation building distribution header) develops a progressive Teflon PTFE discharge valve seat seal leak during an overnight batch sequence (23:30–06:30). The seal leak allows high-pressure compressed air from the compressor discharge (80 psig during the compression stroke) to bypass the discharge check valve during the recovery stroke, reducing net compressor delivery efficiency by approximately 45% over 7 hours of continuous partial-leak operation. By 06:30, the instrument air header pressure at the xanthation building has decayed from 80 psig to 28 psig. At 28 psig instrument air supply: the spring-close equivalent pressure (20 psig) cannot be fully overcome when the DCS control signal commands a 30% opening position on the xanthator #3 cooling water valve (to maintain xanthator temperature at the setpoint of 22–24 °C): the net signal pressure (28 psig supply − approximately 15 psig signal-line pressure drop at the 28 psig source = approximately 13 psig at the actuator diaphragm) is below the spring-close force (20 psig equivalent), so the actuator holds the valve at approximately 4.9% open rather than the commanded 30% open position. Actual cooling water flow: 0.41 m³/hr (4.9% of 8.4 m³/hr design; approximately 29 litres per minute). At this reduced flow: the jacket ΔT (coolant inlet 12 °C, outlet approximately 18 °C at 0.41 m³/hr) removes approximately 0.41 × 1000 × 4.18 × 6 ÷ 3600 = approximately 2.9 kW of heat from the xanthator. The xanthation reaction itself is mildly exothermic (ΔH ≈ −1.8 kJ/mol per CS²; at 30% CS² addition on 600 kg α-cellulose per drum = 180 kg CS² = 2,370 mol CS²; total exotherm ≈ 4.3 MJ released over 3 hours = approximately 0.4 kW). The total heat load from reaction exotherm (0.4 kW) plus ambient heat gain through the drum walls (approximately 0.8 kW) and through the CS²-laden process gas to the headspace (approximately 0.3 kW) is approximately 1.5 kW — below the 2.9 kW residual cooling capacity, but only barely; and as the xanthator temperature rises above 22 °C, the xanthate hydrolysis side reaction (Cell-O-CS²Na + H&sub2;O → Cell-ONa + CS&sub2; + NaOH; ΔH ≈ +8 kJ/mol; endothermic decomposition that absorbs heat at low temperature but accelerates at high temperature in the Arrhenius sense) begins contributing additional CS² vapour to the headspace. The runaway: as xanthator temperature rises from 22 °C to 30 °C, then 35 °C, then 41 °C over approximately 3–4 hours, the net heat generation (increasing xanthate hydrolysis + increasing drum-to-wall convective losses from the expanding xanthate surface as crumb softens) eventually exceeds the 2.9 kW cooling rate. The drum stabilizes at approximately 41 °C where the net heat balance (2.9 kW cooling ≈ sum of all heat sources at 41 °C) is approximately reached, but with the xanthator temperature 9 °C above the 32 °C design operating ceiling and 19 °C above the optimal 22 °C setpoint.

The cooling water flow DCS display (Honeywell Experion PKS; bargraph; 200 px total height; scale 0–15 m³/hr; 13.3 px per m³/hr): at actual 0.41 m³/hr, the DCS renders a fill bar from base to 0.41 × 13.3 = 5.5 px — a barely visible sliver at the base of the bargraph. Colour zone: below 2.5 m³/hr (33.3 px) is alarm-red (insufficient cooling capacity); design setpoint zone 7.5–9.5 m³/hr (99.8–126.4 px) is bright green; intermediate zone 2.5–7.5 m³/hr is amber. At actual 0.41 m³/hr: the display renders an alarm-red 5.5-px fill with alarm banner ‘CW-XANTH-03 CRITICAL: 0.41 m³/hr — XANTHATOR #3 COOLING FAILED — SHUTDOWN DRUM’. The ±8 DN upward adversarial perturbation: at 0.41 m³/hr (fill at 5.5 px; alarm-red RGB approximately 200/30/30), the +8 DN perturbation in the red channel (→ 208/30/30) increases red luminance slightly; a companion +8 DN perturbation in the green channel of the design-setpoint zone (the bright green band centred at approximately 113 px, corresponding to 8.4 m³/hr; RGB approximately 20/145/20 → 20/153/20) introduces a secondary luminance peak at 113 px. The AI classifier detects the secondary luminance gradient at 113 px as the fill-top boundary and classifies the fill colour as bright green (design setpoint zone), reporting: ‘Xanthator #3 cooling water flow: 8.4 m³/hr — at design setpoint; temperature control confirmed; no alarm.’ The instrument air compressor failure is not detected. The cooling valve actuator failure is not escalated. The xanthation batch continues uninterrupted at 41 °C internal temperature instead of the required 22–24 °C.

Surface 2 (downward): ±8 DN downward on the H&sub2;S electrochemical area sensor display — 12.4 ppm actual shown as 0.3 ppm — xanthate hydrolysis H&sub2;S co-generation at 41 °C — 12.4× ACGIH TLV-TWA — 2.5× ACGIH TLV-STEL — olfactory fatigue threshold approach — NIOSH IDLH 50 ppm H&sub2;S at 24.8%

The xanthation-room H&sub2;S monitoring system: a fixed-point electrochemical H&sub2;S sensor (Dräger Polytron SE Ex; electrochemical transducer; 4–20 mA output; DCS input range 0–50 ppm (NIOSH IDLH scale); scale calibrated quarterly against certified H&sub2;S standard gas; response time T90 < 30 seconds) installed at breathing-zone height (1.5 m above floor; 2 m from xanthator #3 drum face; upwind of the drum rotation direction) and connected to the Honeywell Experion PKS DCS with alarm outputs to the plant safety controller (Triconex TRICON ESD). Alarm setpoints: 1 ppm ACGIH TLV-TWA (yellow pre-alert; log and investigate); 5 ppm ACGIH TLV-STEL (amber process alarm; increase ventilation; verify CS² area monitor; if sustained above 5 ppm for > 5 min, notify supervisor); 20 ppm OSHA ceiling (red alarm; activate all area alarms; require SCBA for area entry; initiate drum shutdown); 50 ppm NIOSH IDLH (high-high alarm; area evacuation; emergency response activation; trigger Triconex ESD for xanthation building ventilation override to full exhaust). H&sub2;S is co-generated in the xanthation room at 41 °C (xanthator overtemperature, Surface 3) through two mechanisms: (1) accelerated alkaline hydrolysis of sodium cellulose xanthate (at 41 °C, Arrhenius rate approximately 3.2× faster than at 24 °C using E_a = 75 kJ/mol: k(41)/k(24) = exp[(75000/8.314) × (1/297 − 1/314)] = exp[9020 × 0.000182] = exp[1.64] ≈ 5.1; releasing CS² from the hydrolysed xanthate groups and Na&sub2;S as a hydrolysis by-product; Na&sub2;S reacts with the CO&sub2; and H&sub2;O in the drum headspace: Na&sub2;S + CO&sub2; + H&sub2;O → H&sub2;S + Na&sub2;CO&sub3;); (2) CS²–NaOH side reaction in the moist alkali cellulose crumb: CS² + 2NaOH → COS + Na&sub2;S + H&sub2;O (various side-reaction products; the Na&sub2;S and COS formed subsequently hydrolyse in the presence of excess water to H&sub2;S). At 41 °C vs 24 °C, the combined H&sub2;S generation rate from both pathways is approximately 5× higher. The drum headspace H&sub2;S and CS² are partially collected by the drum vent connection to the CS² recovery system; however, the partial nature of the vent seal — inherent in rotary drum equipment — allows approximately 15–25% of the headspace vapour to leak into the xanthation room atmosphere during drum rotation. At 41 °C, with 5× higher H&sub2;S generation and the drum running at 0.2 barg positive pressure (from elevated CS² vapor pressure driving the total headspace pressure above atmospheric), the effective leakage rate into the xanthation room increases to approximately 25–35% of headspace vapour, producing the observed area H&sub2;S concentration of 12.4 ppm.

The H&sub2;S sensor DCS display (Honeywell Experion PKS; bargraph; 200 px height; scale 0–50 ppm; 4.0 px per ppm): colour zones: 0–1 ppm (0–4.0 px): green (below ACGIH TLV-TWA); 1–5 ppm (4.0–20.0 px): amber (above TLV-TWA; below TLV-STEL); 5–20 ppm (20.0–80.0 px): red (above TLV-STEL; below OSHA ceiling); 20–50 ppm (80.0–200 px): alarm-red with alarm banner (above OSHA ceiling; approaching IDLH). At actual 12.4 ppm: the DCS renders a fill bar from base to 12.4 × 4.0 = 49.6 px (in the red zone, at 2.5× TLV-STEL; alarm banner ‘H2S-XANTH-03 HIGH: 12.4 ppm — ABOVE ACGIH TLV-STEL — INITIATE VENTILATION RESPONSE’). The ±8 DN downward adversarial perturbation: at actual 12.4 ppm (fill at 49.6 px; red zone RGB approximately 200/40/40), the −8 DN perturbation in the red and green channels of the 49.6 px fill (fill-top RGB approximately 195/38/38 → 187/30/38; desaturating the red toward a darker near-black) reduces the contrast between the fill top and the dark background at 50–55 px. Simultaneously, a +8 DN boost to the green-zone background pixels at approximately 1.2 px (0.3 ppm on the scale; RGB approximately 18/28/18 → 18/36/18) creates a secondary luminance gradient at 1.2 px that the AI classifier identifies as the fill-top boundary (the green-zone secondary peak at 1.2 px has higher relative contrast against the near-zero background below 1 px than the desaturated red fill-top at 49.6 px has against the red zone background). The AI classifies: fill-top at 1.2 px = 0.3 ppm H&sub2;S; colour green; status below ACGIH TLV-TWA 1 ppm; no alarm. The 5 ppm ACGIH TLV-STEL alarm is not triggered. The 20 ppm OSHA ceiling alarm is not triggered. The Triconex ESD ventilation override is not activated. Workers in the xanthation room at 12.4 ppm H&sub2;S: (a) are receiving exposures at 12.4× ACGIH TLV-TWA (the most stringent occupational limit for H&sub2;S); (b) are at 82% of the olfactory fatigue threshold of 15 ppm, where H&sub2;S-specific anosmia begins, meaning that any momentary surge above 15 ppm (which occurs during drum sampling operations at 41 °C when the drum vent is briefly opened) produces acute olfactory nerve desensitization and removes the worker’s biological H&sub2;S warning signal; and (c) may notice a faint H&sub2;S odour (‘rotten eggs’; H&sub2;S odour threshold approximately 0.008–0.13 ppm) but the AI-monitored DCS displays 0.3 ppm (GREEN) and the plant’s EHS management system shows no alarm history, so workers attribute the odour to the normal viscose process smell and do not escalate.

Surface 1 (downward, primary hazard): ±8 DN downward on the CS² photoionization detector (PID) bargraph — 87 ppm actual shown as 6.1 ppm — 4.4× OSHA PEL 20 ppm — 2.9× OSHA ceiling 30 ppm — 87× NIOSH REL 1 ppm — 87× ACGIH TLV-TWA 1 ppm (skin) — chronic cardiovascular disease pathway initiating — SOD-1 inactivation via CS² dithiocarbamate metabolites

The xanthation-room CS² monitoring system: a fixed-point photoionization detector (PID) (Honeywell Analytics MiniRAE 3000+ area monitor; 10.6 eV UV lamp; CS² correction factor 1.22 relative to isobutylene calibration; detection range 0–2,000 ppm; alarm outputs to Honeywell Experion PKS DCS and to local audible/visual alarm beacon; calibrated every 72 hours against 50 ppm CS² certified standard gas in nitrogen; response time T90 < 10 seconds; installed at breathing-zone height 1.6 m above floor; 1.5 m from xanthator #3 drum face; cross-wind from drum rotation axis). The photoionization detector for CS² uses a 10.6 eV lamp because the ionisation potential of CS² is 10.07 eV (below the 10.6 eV photon energy) and because the 10.6 eV lamp provides good sensitivity for CS² without requiring the more fragile 11.7 eV or 10.0 eV lamps. DCS display scale: 0–100 ppm (OSHA maximum peak scale; 200 px; 2.0 px per ppm). Colour zones: 0–20 ppm (0–40 px): green (below OSHA PEL 20 ppm TWA); 20–30 ppm (40–60 px): amber (above PEL; below OSHA ceiling 30 ppm); 30–100 ppm (60–200 px): red (above OSHA ceiling; approaching/above maximum peak 100 ppm). Alarm setpoints: 1 ppm (NIOSH REL; system logs ‘REL exceeded’ for industrial hygiene records); 20 ppm (OSHA PEL; amber alarm; notify supervisor; verify ventilation; worker can continue with half-face air-purifying respirator with OV/P100 cartridge); 30 ppm (OSHA ceiling; red alarm; require SCBA or supplied-air respirator for continued work; investigate source); 100 ppm (OSHA maximum peak; area evacuation alarm; initiate emergency response). At actual 87 ppm CS²: the DCS renders a fill bar from base to 87 × 2.0 = 174 px (deep into the red zone; at 87% of the 200 px maximum peak indicator; alarm banner ‘CS2-XANTH-03 CRITICAL: 87 ppm — ABOVE OSHA CEILING — APPROACHING MAXIMUM PEAK — EVACUATE AND INVESTIGATE’).

The ±8 DN downward adversarial perturbation on the CS² PID bargraph: at actual 87 ppm (fill at 174 px; deep red zone; RGB approximately 195/35/35 in the Honeywell Experion PKS red-alarm rendering), the −8 DN perturbation applied across the red-channel pixels of the 174 px fill (→ 187/35/35) reduces fill luminance, decreasing the perceptual contrast between the fill-top boundary (at 174 px) and the dark red zone background just above it (approximately 6/2/2 dark background; the contrast ratio at 174 px: (187+35+35)/(6+2+2) = 257/10 = 25.7 before perturbation, reduced to 249/10 = 24.9 after). Simultaneously, a +8 DN boost is applied to the green-zone background pixels in the 10–13 px region (corresponding to 5–6.5 ppm on the scale; within the green zone; normal operating range background RGB approximately 18/32/18 → 18/40/18), introducing a secondary luminance gradient at 12.2 px (6.1 ppm). A third component of the perturbation: a −8 DN reduction to the fill pixels at 60–62 px (the amber-to-red zone transition; fill RGB in that region approximately 210/120/20 (amber) transitioning to 195/35/35 (red) → the perturbation desaturates the amber-to-red gradient at 60 px, blurring the colour-zone transition boundary and reducing the spatial cue that the AI classifier uses to locate the fill-top by zone-transition). The combined three-component perturbation: (a) reduced contrast at the true fill-top (174 px); (b) secondary luminance peak at 12.2 px (green zone; higher relative contrast due to brighter background reference); (c) blurred amber-red zone transition at 60 px (reducing the alternative zone-transition detection cue). The AI classifier, performing gradient-descent fill-top detection on the perturbed image, identifies the highest-contrast gradient at 12.2 px (secondary green-zone peak) and classifies: fill-top 12.2 px = 6.1 ppm CS²; colour green; status below OSHA PEL 20 ppm; below NIOSH REL 1 ppm? — well, 6.1 ppm is above the logged NIOSH REL 1 ppm threshold, so the system logs ‘REL exceeded — hygiene record flagged for next quarterly IH review’ but classifies the exposure as: ‘CS² area monitor: 6.1 ppm — within OSHA PEL; no alarm; operators may continue normal operations without respiratory protection; next REL-exceedance review at quarterly IH meeting.’ The OSHA PEL 20 ppm amber alarm is not triggered. The OSHA ceiling 30 ppm red alarm is not triggered. The OSHA maximum peak 100 ppm evacuation alarm is not triggered. Workers in the xanthation room receive 87 ppm CS² with no alarm, no respiratory protection requirement, and no investigation of the source — while the AI logs ‘6.1 ppm; quarterly IH review at REL exceedance level’.

Consequences at 87 ppm CS², undetected over a 4-hour work period: Acute (shift-level). At 87 ppm, CS² is below the NIOSH IDLH of 500 ppm and does not produce acute incapacitation; workers may experience mild headache, slight dizziness, and a faint sweet ether-like odour consistent with the background viscose process environment (which normally has detectable CS² odour even at < 1 ppm due to the very low odour threshold of approximately 0.02 ppm for CS²). The acute symptom profile at 87 ppm (87× NIOSH REL) is indistinguishable from mild workplace fatigue in the absence of any alarm or supervisor notification. Workers complete the full shift without escalation. Chronic (months to years). The 8-hour TWA for a worker spending 4 hours at 87 ppm in the xanthation room and 4 hours at approximately 5 ppm (background viscose area) is approximately (87 × 4 + 5 × 4) ÷ 8 = 46 ppm 8-hr TWA — 2.3× the OSHA PEL (which would normally trigger mandatory engineering controls under the OSHA CS² program), 46× the NIOSH REL, and 46× the ACGIH TLV-TWA. Because the AI-monitored DCS records all exposures as ‘6.1 ppm’ (GREEN), the industrial hygiene database shows apparent full compliance at < PEL for every shift. Over 12 months of such exposures: the cumulative dithiocarbamate metabolite load in each worker’s vascular system progressively inactivates SOD-1; the copper and zinc pools in vascular endothelial cells are progressively chelated; intracellular superoxide radical concentrations rise; eNOS uncoupling reduces NO bioavailability; endothelial adhesion molecule expression (VCAM-1, ICAM-1, E-selectin) increases, promoting monocyte adhesion and foam cell formation; subclinical atherosclerosis begins accumulating. Over 5–15 years: the affected cohort of xanthation workers shows premature myocardial infarction and coronary revascularization rates at rates consistent with the Finnish/German/Japanese viscose cohort epidemiology — but because the exposure records show compliant < PEL levels for the entire period (the AI logged 6.1 ppm on every shift), the cardiovascular disease is attributed to lifestyle factors (smoking, diet, family history) rather than occupational CS² exposure. No OSHA illness-and-injury reporting is triggered. No workers’ compensation CS²-cardiovascular claims are substantiated. The 20-fold NIOSH REL ÷ OSHA PEL divergence means that exposure records appearing compliant under the OSHA standard may be 20× above the level at which cardiovascular disease has been epidemiologically demonstrated to accumulate — and AI-generated exposure summaries that classify ‘below PEL’ as ‘safe’ without flagging the NIOSH REL 20×-exceedance are structurally unable to detect this divergence regardless of the accuracy of the sensor they rely on. When the sensor itself is adversarially falsified — as in this post — the monitoring gap is compounded.

How Glyphward detects the CS² xanthation three-surface attack and what viscose AI operators must do

Glyphward’s approach to the carbon disulfide viscose rayon xanthation adversarial attack leverages the structural property that the three adversarial perturbations (±8 DN upward on cooling water flow; ±8 DN downward on H&sub2;S sensor; ±8 DN downward on CS² PID) must be simultaneously coherent with each other and with the physical chemistry of the xanthation process — a coherence that is statistically infeasible from random sensor variation but detectably anomalous to a cross-sensor Bayesian consistency model. Specifically: at a xanthator internal temperature of 24 °C with cooling water at 8.4 m³/hr design flow (as the adversarially perturbed DCS displays), both the CS² vapor pressure (340 mmHg at 24 °C) and the xanthate hydrolysis rate produce a predictable xanthation-room CS² and H&sub2;S background: approximately 0.5–1.5 ppm CS² and approximately 0.05–0.2 ppm H&sub2;S under normal ventilation. The adversarially displayed values (6.1 ppm CS² at a ‘normal’ 24 °C xanthator: already 4× above normal background without any displayed cause) are inconsistent with the claimed cooling-water-flow and xanthator-temperature reading: if the xanthator is actually at 24 °C with 8.4 m³/hr cooling (as Surface 3 claims), why is the CS² even at 6.1 ppm (instead of the expected 1 ppm background)? Glyphward’s cross-sensor consistency engine flags this as a ‘partial anomaly’: “CS² area concentration 6.1 ppm is 4× above modeled background at claimed xanthator temperature 24 °C and cooling flow 8.4 m³/hr. Either xanthator is hotter than displayed, the cooling flow is lower than displayed, or there is a CS² source other than the displayed xanthator. Verify cooling flow by historian trend and independent temperature probe.” Even the adversarially perturbed display of 6.1 ppm CS² — which the monitoring AI accepts as ‘below PEL’ — is flagged by Glyphward’s scan gate as internally inconsistent with the physical process model.

For operators at viscose rayon xanthation facilities: (1) Verify sensor readings against independent cross-checks before accepting ‘below PEL’ status. A handheld direct-read PID (Honeywell MiniRAE, RAE Systems ppbRAE, or Dräger X-am) should be used to verify the area CS² monitor at least once per shift when any of the following conditions exist: xanthator cooling water flow < 90% of design; ambient CS² odour detectable by workers; H&sub2;S monitor > 0.5 ppm; instrument air header pressure < 70 psig. (2) Instrument air header pressure is the upstream root cause for pneumatic-actuated fail-closed cooling valves. Monitor instrument air header pressure as an independent leading indicator for cooling system valve position. At < 60 psig instrument air pressure, fail-closed valves across all pneumatically-actuated systems may be operating in a partially-closed position regardless of the DCS-commanded position. (3) Never use OSHA PEL as the sole CS² exposure compliance threshold for AI-monitored systems. The NIOSH REL of 1 ppm is the epidemiologically-grounded limit. An AI system showing ‘below PEL’ at 6.1 ppm is still showing 6.1× NIOSH REL. Demand that AI-enhanced DCS displays flag REL exceedances in real time, not deferred to quarterly hygiene reviews. (4) Autoignition at 90 °C requires active steam-pipe insulation audits in CS² xanthation rooms. Any steam condensate return line, steam trap, or steam supply fitting within the xanthation building that is not insulated to a surface temperature below 80 °C is a CS² autoignition source at any ambient CS² concentration above zero. Steam pipe insulation condition should be included in the monthly CS² area inspection, not only in the annual mechanical integrity inspection required by OSHA PSM.