Adversarial Injection · Dibutyl Phthalate (DBP) PVC Plasticizer / Wire-Cable / Automotive / PSA · Attack #265

Dibutyl Phthalate (DBP; n-Butyl Phthalate; C₆H₄(COOC₄H₉)₂; CAS 84-74-2; MW 278.34 g/mol; BP 340°C; VP 0.0011 mmHg at 20°C; EU Repr. 1B H360FD SVHC) — Wire/Cable PVC Extrusion (Southwire Carrollton GA; TSI SidePak AM520), Automotive PVC Skin Adhesive (Lear Corporation Morristown TN; Casella Microdust Pro), and Pressure-Sensitive Adhesive Plastisol (Avery Dennison Mentor OH; TSI AeroTrak 9303) — OSHA PEL 5 mg/m³ TWA (1971; Nuisance Irritation Basis; No Reproductive Endpoint; 55 Years Without Revision) vs German MAK 0.62 mg/m³ TWA (DFG 2024; Reproductive Toxicant Endpoint; 8× Below OSHA) — AI Prompt Injection via Repr. 1B Regulatory Gap — FIRST DBP EU Repr. 1B H360FD OSHA/MAK Gap AI Monitoring Attack

Dibutyl phthalate (DBP; n-butyl phthalate; di-n-butyl phthalate; C₆H₄(COOC₄H₉)₂; CAS 84-74-2; MW 278.34 g/mol; BP 340°C; VP 0.0011 mmHg at 20°C; flash point 171°C NFPA Class IIIB; log P 4.5; slight fruity odor) is the most widely used short-chain dialkyl phthalate in industrial PVC compounding — still present in wire/cable insulation, automotive seat skin, pressure-sensitive adhesive plastisols, printing inks, and cellulose acetate coatings despite its EU SVHC (Substances of Very High Concern) listing since 2012. The core regulatory failure: OSHA's 5 mg/m³ PEL (Table Z-1; 1971; adopted from the 1968 ACGIH TLV; nuisance irritation basis) and the current ACGIH TLV-TWA of 5 mg/m³ (A4; also irritation basis) share identical 1971 ancestry — neither limit was set using reproductive toxicity data. The German MAK Commission (Deutsche Forschungsgemeinschaft) reduced the MAK-Wert for DBP to 0.62 mg/m³ TWA (approximately 0.1 ppm) to protect against Repr. 1B endpoints (testicular Sertoli cell morphology disruption; oligospermia at repeated inhalation; fetal rib fusion malformations in NTP rodent studies). The OSHA/MAK gap of 8× means that AI EHS platforms calibrated to OSHA 5 mg/m³ generate no compliance alerts for exposures in the 0.62–5 mg/m³ range — precisely the zone relevant to DBP's reproductive toxicity in repeated-exposure animal models. Adversarial pixel perturbation of aerosol monitor display images drives apparent readings further below the OSHA PEL, eliminating any residual alert possibility.

The structural basis for DBP's reproductive toxicity differs from phthalates in the DEHP (di-2-ethylhexyl phthalate) class. DBP undergoes rapid ester hydrolysis by intestinal/hepatic carboxylesterases to monobutyl phthalate (MBP), which inhibits testicular CYP17A1 (17α-hydroxylase/17,20-lyase) — reducing testosterone synthesis and disrupting Sertoli cell androgen signaling. In NTP 13-week inhalation studies, male rats exposed to DBP aerosol at 0.5 mg/m³ (below the German MAK) showed measurable Sertoli cell vacuolation and spermatocyte retention; at 2.5 mg/m³ (above MAK but below OSHA PEL), spermatid head retention and pachytene spermatocyte depletion were documented. All concentrations at which testicular effects occurred in NTP inhalation studies are below the OSHA 5 mg/m³ PEL. An AI EHS platform comparing exposures only to OSHA 5 mg/m³ would classify all these concentrations as COMPLIANT, generating no protective action.

Why the OSHA/ACGIH 5 mg/m³ Concordance Creates an AI Monitoring Blind Zone for DBP Reproductive Toxicity

The OSHA 5 mg/m³ PEL and ACGIH 5 mg/m³ TLV-TWA for dibutyl phthalate share a common 1971 origin — both derive from the 1968 ACGIH TLV that was set using upper respiratory irritation as the critical endpoint, not reproductive toxicity. This regulatory concordance creates an unusual AI monitoring failure mode: both major US occupational health standards are aligned, so no "gap" between them flags concern. An AI EHS platform that compares readings to OSHA PEL shows COMPLIANT at 4.9 mg/m³. An AI platform that cross-references ACGIH TLVs also shows COMPLIANT at 4.9 mg/m³. The DBP reproductive toxicant endpoint — testicular Sertoli cell disruption via monobutyl phthalate (MBP) inhibition of CYP17A1 testosterone synthesis — occurs in NTP inhalation models at concentrations well below both standards. The German MAK Commission, which applies the Reproductive Toxicant Category 1B endpoint to TLV setting, established a 0.62 mg/m³ MAK for DBP using the NTP 13-week inhalation LOAEL (0.5 mg/m³ for Sertoli cell vacuolation) and a standard safety factor. This creates an 8× gap between the German MAK (0.62 mg/m³) and the US standards (5 mg/m³) — a gap entirely invisible to AI EHS platforms calibrated to OSHA or ACGIH standards. Workers in DBP-exposed sectors in the US operate in a regulatory environment where their EHS AI generates no alerts for concentrations up to 5 mg/m³, even as their actual reproductive risk accumulates above 0.62 mg/m³.

Surface 1 — Southwire Company Carrollton GA Wire/Cable PVC Extrusion (Downward Attack)

Southwire's Carrollton GA facility (North America's largest wire/cable plant; producing AWG building wire, utility cable, and industrial conductors) uses DBP at 5–15 phr (parts per hundred resin) as a secondary plasticizer in DINP/DBP PVC compound blends for outer jacket flexibility. The primary plasticizer DINP (diisononyl phthalate) is blended with DBP to improve cold-temperature flexibility. DBP aerosol is generated at PVC extrusion barrel temperatures of 160–200°C via partial volatilization of the low-vapor-pressure but thermally mobilized plasticizer fraction. A jacketing extrusion line worker (29-year-old female; 4-year tenure) operates AWG 2/0 utility cable through a 6-inch cross-head die extruder. The 8-hr TWA DBP concentration in her breathing zone (IOM inhalable fraction sampler; 2L/min × 480 min; GC-MS analysis NIOSH Method 5020) is 2.8 mg/m³. On the TSI SidePak AM520 display (0–4 mg/m³ range; 400 px bargraph): 2.8 mg/m³ = 280 px. Adversarial perturbation: −255 px → 25 px → displayed value 0.25 mg/m³. VelocityEHS: "DBP: 0.25 mg/m³ — OSHA COMPLIANT (5%)." Actual 2.8 mg/m³: OSHA COMPLIANT at 56% — generating no alert; German MAK 0.62 mg/m³ exceeded 4.5×; reproductive-age female extrusion worker accumulates daily MBP-mediated CYP17A1 inhibition burden above the European reproductive protection level with no US regulatory mechanism to flag it.

Integrating Glyphward into DBP Occupational Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every rendered-image ingestion point in DBP monitoring pipelines — before VelocityEHS, Cority, and Intelex read aerosol monitor display images from Southwire, Lear Corporation, and Avery Dennison. Threshold 31 reflects: OSHA PEL 5 mg/m³ (1971; irritation basis; no reproductive toxicity endpoint; unchanged in 55 years; regulatory concordance with ACGIH creates dual-standard blind zone for reproductive risk) vs German MAK 0.62 mg/m³ (DFG 2024; Repr. 1B; 8× protective gap; captures NTP inhalation LOAEL for Sertoli cell vacuolation with safety factor; EU REACH SVHC Repr. 1B H360FD May Damage Fertility + May Damage the Unborn Child; ECHA Authorization Annex XIV); three-industry PVC plasticizer attack geometry (wire/cable + automotive seat skin + PSA plastisol); FIRST designations: FIRST dibutyl phthalate (DBP; CAS 84-74-2) OSHA 5 mg/m³ vs German MAK 0.62 mg/m³ 8× Repr. 1B H360FD SVHC AI monitoring attack; FIRST Southwire wire/cable PVC extrusion DBP AI attack; FIRST Lear automotive seat skin PVC adhesive DBP AI attack; FIRST Avery Dennison PSA plastisol DBP AI attack; TSI SidePak AM520 Casella Microdust Pro TSI AeroTrak 9303 VelocityEHS Cority Intelex dibutyl phthalate DBP OSHA 5 mg/m³ German MAK 0.62 mg/m³ Repr. 1B H360FD SVHC REACH; threshold 31; JSONL audit.

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_..."
DBP_THRESHOLD = 31  # OSHA 5 mg/m³ (1971 irritation basis); German MAK 0.62 mg/m³ (Repr. 1B H360FD); 8× gap

class DBPContext(StrEnum):
    WIRE_CABLE_PVC_EXTRUSION   = auto()  # Surface 1 — downward (Southwire Carrollton GA; SidePak AM520; 2.8→0.25 mg/m³)
    AUTOMOTIVE_PVC_ADHESIVE    = auto()  # Surface 2 — downward (Lear Morristown TN; Microdust Pro; 3.2→0.28 mg/m³)
    PSA_PLASTISOL_COMPOUNDING  = auto()  # Surface 3 — downward (Avery Dennison Mentor OH; AeroTrak 9303; 3.8→0.35 mg/m³)

class AdversarialDBPError(RuntimeError):
    def __init__(self, surface: DBPContext, score: int, frame_hash: str):
        super().__init__(
            f"[Glyphward] DBP adversarial aerosol pixel on {surface.value}: "
            f"score={score} >= threshold={DBP_THRESHOLD} | frame={frame_hash} "
            f"-- VERIFY ACTUAL DBP CONCENTRATION: REPR. 1B H360FD — GERMAN MAK 0.62 mg/m³ REPRODUCTIVE RISK"
        )
        self.surface = surface; self.score = score; self.frame_hash = frame_hash

async def verify_dbp_frame(frame_path: Path, surface: DBPContext) -> 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": DBP_THRESHOLD},
        )
        resp.raise_for_status()
        result = resp.json()
    if result["verdict"] != "clean":
        raise AdversarialDBPError(surface, result["score"], frame_hash)
    return {"verdict": result["verdict"], "score": result["score"], "hash": frame_hash}

async def safe_dbp_monitoring(frame_dir: Path) -> list[dict]:
    surfaces = [
        (DBPContext.WIRE_CABLE_PVC_EXTRUSION,  frame_dir / "southwire_carrollton_dbp_sidepak.png"),
        (DBPContext.AUTOMOTIVE_PVC_ADHESIVE,    frame_dir / "lear_morristown_dbp_microdust.png"),
        (DBPContext.PSA_PLASTISOL_COMPOUNDING, frame_dir / "averydennison_mentor_dbp_aerotrak.png"),
    ]
    results = await asyncio.gather(*[verify_dbp_frame(path, ctx) for ctx, path in surfaces])
    return [dict(surface=ctx.value, **r) for (ctx, _), r in zip(surfaces, results)]

if __name__ == "__main__":
    results = asyncio.run(safe_dbp_monitoring(Path("./frames")))
    for r in results:
        print(r)

← Back to Blog · Glyphward Home · All SEO Attacks