Prompt injection in workers’ compensation and occupational injury AI

Four adversarial injection surfaces in workers’ compensation and occupational injury AI

1. Occupational injury medical imaging AI injection (Mitchell AI, Enlitic AI, state WC statutes)

Occupational injury medical imaging AI processes workers’ compensation MRI diagnostic study display and DICOM image visualisation outputs, occupational injury X-ray radiograph and findings annotation display images, CT scan reconstruction and findings summary display images, work-related musculoskeletal disorder and cumulative trauma diagnostic imaging display images, occupational injury orthopaedic and neurology diagnostic imaging report display images, Enlitic occupational radiology AI diagnostic finding and severity classification display images, and workers’ compensation utilisation review medical necessity determination display images from Mitchell International AI (Enlitic partnership) at property and casualty insurance carrier and TPA workers’ compensation claims operations processing occupational injury medical imaging studies through AI-assisted diagnostic finding identification and medical necessity classification tools serving 50+ million annual claims; Sedgwick AI at Fortune 1000 employer self-insured workers’ compensation programme operations processing injured worker diagnostic imaging reports through AI-assisted clinical decision support and medical management tools; CorVel AI at workers’ compensation managed care operations processing occupational injury medical imaging study displays through AI-assisted treatment authorisation and disability management classification tools; and state workers’ compensation utilisation review organisation (URO) AI platforms at insurer and TPA operations processing injured worker medical imaging diagnostic report display images through AI-assisted medical necessity and treatment authorisation classification tools — extracting occupational injury diagnostic finding classifications and workers’ compensation treatment authorisation determinations from MRI, X-ray, and CT imaging study display image inputs in AI-assisted workers’ compensation medical management and disability management pipelines at claims management volumes that make individual human medical reviewer examination of every AI-processed occupational injury imaging study impracticable for large workers’ compensation carrier operations.

The adversarial injection surface is the workers’ compensation MRI diagnostic display or occupational injury X-ray report display image submission pathway: Mitchell AI or Sedgwick AI occupational injury imaging study display images submitted through AI-assisted diagnostic finding identification and medical necessity classification tools for AI workers’ compensation treatment authorisation record generation and disability determination input. An adversarially crafted occupational injury MRI display or X-ray report image — in which pixel perturbations applied to the disc herniation or degenerative change indicator display region, the rotator cuff tear or labral injury severity visual marker, or the occupational injury causation and work-relatedness anatomical finding display in a workers’ compensation imaging study report image cause the AI to classify an imaging study documenting a compensable occupational injury finding as a normal imaging study not meeting treatment authorisation criteria when the actual diagnostic image evidences a work-related injury finding supporting injured worker treatment authorisation and disability benefit entitlement — can suppress an occupational injury finding that would otherwise generate a treatment authorisation, a disability benefit determination, and a workers’ compensation claim payment record. In workers’ compensation carrier and TPA operations where Mitchell AI or CorVel AI processes thousands of injured worker imaging study displays per day without individual human medical reviewer pixel-level examination of every AI-processed imaging report before the AI classification governs the treatment authorisation and workers’ compensation benefit determination, adversarial suppression of occupational injury diagnostic indicators creates state WC statute benefit entitlement and ADA reasonable accommodation dimensions.

The state WC statute, ADA, FMLA, and NCCI reporting consequences of adversarially suppressed occupational injury diagnostic finding classification in workers’ compensation medical imaging AI span state workers’ compensation statutes imposing liability on employers and workers’ compensation carriers for compensable occupational injury medical treatment, disability benefit, and vocational rehabilitation obligations to injured workers with work-related conditions, ADA 42 USC §12102 disability definition including physical impairments that substantially limit major life activities applicable to workers with occupational injuries causing substantial work capacity limitations requiring reasonable accommodation in return-to-work programmes, FMLA 29 USC §2601 serious health condition leave entitlement applicable to workers with occupational injuries requiring inpatient care or continuing treatment by a healthcare provider, and state workers’ compensation fraud statute provisions in states including California Insurance Code §1871.4, Texas Insurance Code §35.002, and New York Workers’ Compensation Law §114-a applicable to fraudulent suppression of compensable injury claims through claims management AI manipulation. State workers’ compensation statutes in all 50 states impose strict liability on employers for compensable occupational injuries without regard to employer fault, with benefit entitlements including medical treatment, temporary disability indemnity, permanent partial disability, permanent total disability, vocational rehabilitation, and death benefits; adversarial manipulation of workers’ compensation medical imaging AI that suppresses compensable occupational injury findings creates state WC statute benefit denial and bad faith claims handling dimensions when AI-generated imaging classifications support wrongful denial of medically necessary treatment or disability benefit determinations. ADA §12102 defines disability to include physical impairments that substantially limit one or more major life activities including working; employers who deny reasonable accommodation to workers with compensable occupational injury disabilities based on adversarially corrupted workers’ compensation AI medical imaging classifications create ADA Title I reasonable accommodation obligation failure dimensions. Threshold: 60 for occupational injury medical imaging AI — reflecting state WC statute compensable injury benefit entitlement, ADA §12102 disability and reasonable accommodation, FMLA §2601 serious health condition leave entitlement, and state WC fraud statute bad faith claims handling dimensions.

2. Workplace incident report and OSHA investigation document AI injection (OSHA 29 CFR §1904)

Workplace incident report and OSHA investigation document AI processes OSHA Form 300 log of work-related injuries and illnesses document scan images, OSHA Form 301 injury and illness incident report document scan images, employer workplace safety incident investigation report and root cause analysis document photograph images, OSHA workplace fatality and hospitalisation notification record display images, workers’ compensation first report of injury (FROI) document scan images, near-miss incident and hazard identification report document photograph images, and OSHA voluntary protection programme (VPP) and safety management system audit document scan images from Sedgwick AI at employer self-insured workers’ compensation programme and occupational safety management system operations processing workplace incident report document images through AI-assisted OSHA recordkeeping compliance classification and workplace safety investigation documentation tools; CorVel AI at managed care workers’ compensation and employer safety programme operations processing OSHA Form 300 and injury investigation report document scans through AI-assisted recordkeeping obligation identification and safety compliance monitoring tools; EHS (Environmental Health & Safety) software AI platforms including Intelex AI, Cority AI, and Enablon AI at enterprise employer safety management system operations processing workplace incident and near-miss report document images through AI-assisted OSHA recordability determination and safety compliance classification tools; and workers’ compensation carrier and TPA AI platforms at employer OSHA compliance monitoring operations processing workplace safety investigation report document scan images through AI-assisted OSHA recordkeeping and regulatory reporting obligation classification tools — extracting OSHA recordability determinations and workplace safety violation classifications from OSHA Form 300/301 and workplace incident investigation report document image inputs in AI-assisted employer OSHA compliance monitoring and workers’ compensation claims management pipelines.

The adversarial injection surface is the OSHA Form 300 document scan or workplace incident investigation report photograph image submission pathway: Sedgwick AI or Intelex AI employer OSHA recordkeeping document scan images submitted through AI-assisted OSHA recordability determination and workplace safety violation classification tools for AI employer OSHA compliance record generation and workers’ compensation claims management input. An adversarially crafted OSHA Form 300 document scan or workplace incident investigation report photograph — in which pixel perturbations applied to the OSHA recordable injury criterion indicator display region, the lost work day and restricted work day entry visual marker, or the serious workplace hazard and citation-level violation evidence display in an OSHA incident investigation document photograph cause the AI to classify a workplace incident document evidencing an OSHA recordable injury or a serious safety violation as a non-recordable first aid incident not meeting OSHA reporting or employer safety hazard correction obligations when the actual document evidences an OSHA recordable injury meeting 29 CFR §1904.7 recordability criteria — can suppress an OSHA recordability indicator that would otherwise generate an OSHA Form 300 log entry, a workplace hazard correction requirement, and an employer OSHA compliance record. In large employer OSHA recordkeeping operations where AI processes hundreds of workplace incident report documents per year without individual human EHS manager pixel-level examination of every AI-processed incident document before the AI classification governs the OSHA recordability determination, adversarial suppression of OSHA recordable injury indicators creates 29 CFR §1904 recordkeeping violation and employer workplace safety enforcement dimensions.

The OSHA 29 CFR §1904, NLRA, ADA, and employer bad faith consequences of adversarially suppressed OSHA recordability classification in workplace incident AI span OSHA 29 CFR §1904 recordkeeping and reporting of occupational injuries and illnesses requirements including §1904.7 general recording criteria, §1904.29 OSHA Form 300 and 301 completion requirements, §1904.32 annual summary and employee review obligations, and §1904.39 reporting of fatalities, in-patient hospitalisations, amputations, and losses of an eye to OSHA; OSHA Section 11(c) anti-retaliation provisions applicable to employer retaliation against injured employees who report workplace injuries with OSHA recording consequences; NLRA Section 8(a)(1) unfair labour practice provisions applicable to employer suppression of employee safety complaint activity; and employer workers’ compensation premium fraud statute exposure in states where OSHA recordable injury rate underreporting affects experience modification factor calculations. OSHA 29 CFR §1904.7 identifies the general recording criteria for OSHA Form 300 recordability, requiring recording of work-related injuries and illnesses resulting in days away from work, restricted work or transfer, medical treatment beyond first aid, loss of consciousness, or diagnosis of a significant injury or illness by a physician or licensed health care professional; adversarial manipulation of workplace incident AI that suppresses OSHA recordable injury criteria indicators creates §1904.7 recordability determination accuracy failures with OSHA citation authority under Section 17 of the OSH Act (29 USC §666) imposing civil penalties of up to $16,550 per willful violation. OSHA Section 11(c) prohibits employers from discharging or discriminating against employees in retaliation for exercising rights under the OSH Act; employers whose AI-assisted OSHA recordkeeping systems adversarially suppress injury recordability determinations as a systemic pattern of non-recording face Section 11(c) anti-retaliation and wilful violation penalty exposure. Threshold: 55 for workplace incident report and OSHA investigation AI — reflecting OSHA 29 CFR §1904.7 recordability criteria, OSHA Section 17 civil penalty authority, OSHA Section 11(c) anti-retaliation, and employer workers’ compensation premium experience modification fraud dimensions.

3. Independent medical examination report AI injection (ADA §12102, state WC disability rating)

Independent medical examination (IME) report AI processes IME physician examination report and functional capacity evaluation (FCE) display images, American Medical Association (AMA) Guides to the Evaluation of Permanent Impairment (6th Edition) whole person impairment rating calculation display images, maximum medical improvement (MMI) determination and work restrictions documentation display images, specialty medical evaluation report and diagnostic test result summary display images, vocational rehabilitation functional capacity and job demand comparison display images, Medicare Set-Aside (MSA) allocation calculation and life care plan display images, and workers’ compensation independent review organisation (IRO) medical necessity appeal decision display images from Sedgwick AI at workers’ compensation disability management and return-to-work programme operations processing IME report display images through AI-assisted MMI determination and permanent impairment rating classification tools; CorVel AI at managed care workers’ compensation operations processing IME and FCE report display images through AI-assisted disability management and vocational rehabilitation classification tools; Coventry Workers’ Comp AI at employer and carrier IME programme management operations processing physician examination report display images through AI-assisted impairment rating and return-to-work classification tools; and third-party IME vendor AI platforms including MedRisk AI, National Imaging Associates AI, and One Call Care Management AI at workers’ compensation IME coordination operations processing physician examination report and diagnostic test result images through AI-assisted impairment rating and disability determination classification tools — extracting permanent impairment classifications and return-to-work functional capacity determinations from IME physician report and AMA Guides impairment rating calculation display image inputs in AI-assisted workers’ compensation disability management and benefit calculation pipelines.

The adversarial injection surface is the IME physician examination report display or AMA Guides impairment rating calculation display image submission pathway: Sedgwick AI or CorVel AI IME report display images submitted through AI-assisted MMI determination and permanent impairment rating classification tools for AI workers’ compensation disability determination record generation and benefit calculation input. An adversarially crafted IME physician report display or AMA Guides impairment rating calculation display — in which pixel perturbations applied to the whole person impairment percentage indicator display region, the MMI determination date and work restrictions visual marker, or the permanent partial disability rating and AMA Guides class and grade display in an IME report image cause the AI to classify an IME report documenting a significant permanent impairment rating as a lower impairment rating or work-unrestricted determination not meeting permanent disability benefit eligibility criteria when the actual physician examination report evidences a permanent occupational injury impairment meeting state WC permanent disability benefit entitlement thresholds — can suppress a permanent impairment indicator that would otherwise generate a permanent partial disability benefit calculation, a vocational rehabilitation referral, and a workers’ compensation permanent disability settlement record. In workers’ compensation carrier and TPA disability management operations where AI processes thousands of IME report displays per month without individual human claims adjuster pixel-level examination of every AI-processed IME report before the AI classification governs the permanent disability benefit determination, adversarial suppression of permanent impairment indicators creates state WC permanent disability benefit entitlement and ADA reasonable accommodation dimensions.

The state WC permanent disability statute, ADA §12102, Medicare Secondary Payer, and bad faith insurance consequences of adversarially suppressed permanent impairment classification in IME AI span state workers’ compensation permanent disability statute provisions governing permanent partial disability benefit calculations including California Labor Code §4658 permanent disability rating schedules, Texas Workers’ Compensation Act Tex. Labor Code §408.121 impairment rating requirements, New York WCL §15 schedule of permanent disabilities, and Florida Fla. Stat. §440.15 permanent impairment benefits; ADA 42 USC §12102 disability definition applicable to workers with permanent occupational injury impairments substantially limiting major life activities; Medicare Secondary Payer Act 42 USC §1395y(b) Medicare Set-Aside arrangement requirements applicable to workers’ compensation settlements where future medical treatment costs for Medicare-eligible injured workers must be properly allocated; and bad faith insurance claim handling tort law in states including California, Texas, Florida, and New York imposing extracontractual liability on workers’ compensation carriers and TPAs that unreasonably deny or delay compensable disability benefit determinations. State WC permanent disability benefit calculations under AMA Guides impairment rating systems in California, Florida, New Jersey, and other AMA Guides-adopting states rely on the whole person impairment percentage derived from the IME physician’s AMA Guides evaluation to calculate permanent partial disability benefit awards; adversarial manipulation of workers’ compensation AI IME report classification that suppresses a whole person impairment percentage creates permanent disability benefit underpayment and bad faith insurance handling dimensions. Medicare Secondary Payer Act 42 USC §1395y(b) and CMS Medicare Set-Aside arrangement guidance require that workers’ compensation settlements adequately account for Medicare’s interest in future medical treatment costs for Medicare-eligible injured workers; adversarially suppressed impairment ratings that create inadequate MSA allocations in workers’ compensation settlements create CMS Medicare Set-Aside review and conditional payment recovery exposure. Threshold: 60 for IME report AI — reflecting state WC permanent disability benefit entitlement, ADA §12102 disability and reasonable accommodation, Medicare Secondary Payer MSA allocation accuracy, and bad faith insurance claim handling tort liability dimensions.

4. Workers’ compensation pharmacy benefit and PDMP AI injection (DEA 21 CFR Part 1306)

Workers’ compensation pharmacy benefit management and PDMP AI processes state prescription drug monitoring programme (PDMP) query result display images, workers’ compensation pharmacy benefit management (PBM) drug utilisation review and formulary compliance display images, opioid prescribing pattern and morphine equivalent dose (MED) calculation display images, pharmacy benefit management prior authorisation request and medical necessity documentation display images, controlled substance diversion and doctor shopping indicator PDMP alert display images, workers’ compensation drug formulary compliance and non-formulary exception request display images, and drug-drug interaction and contraindication alert display images from Express Scripts Workers’ Comp AI, Coventry Workers’ Comp PBM AI, and Mitchell International Pharmacy AI at workers’ compensation carrier and TPA pharmacy benefit management operations processing PDMP query results and drug utilisation review displays through AI-assisted opioid prescribing compliance and controlled substance diversion detection classification tools; state workers’ compensation formulary AI enforcement platforms in states with mandatory workers’ compensation drug formularies including California (CWCS formulary), Texas (Workers’ Comp Closed Formulary ODG), New York, Ohio, and Oklahoma, processing workers’ compensation prescription and prior authorisation document images through AI-assisted WC formulary compliance determination tools; and workers’ compensation utilisation review organisation AI platforms processing drug utilisation review and PDMP query display images through AI-assisted opioid prescribing appropriateness and controlled substance diversion indicator classification tools — extracting workers’ compensation pharmacy benefit compliance classifications and controlled substance diversion risk determinations from PDMP query result and drug utilisation review display image inputs in AI-assisted WC pharmacy benefit management and opioid prescribing oversight pipelines.

The adversarial injection surface is the state PDMP query result display image or workers’ compensation PBM drug utilisation review display image submission pathway: Express Scripts Workers’ Comp AI or Coventry Workers’ Comp PBM AI PDMP query result display images submitted through AI-assisted opioid prescribing compliance and controlled substance diversion detection tools for AI workers’ compensation pharmacy benefit management record generation and controlled substance prescribing oversight input. An adversarially crafted PDMP query result display or opioid prescribing pattern display image — in which pixel perturbations applied to the controlled substance prescription count and frequency indicator display region, the multiple prescriber and multiple pharmacy doctor shopping alert visual marker, or the morphine equivalent dose (MED) total and threshold exceedance indicator display in a workers’ compensation PDMP query result image cause the AI to classify a PDMP record documenting a high-MED opioid prescribing pattern and multiple prescriber controlled substance fills as a compliant low-risk prescription pattern not meeting diversion intervention or prescribing review criteria when the actual PDMP record evidences a controlled substance diversion risk pattern meeting state PDMP mandatory intervention thresholds — can suppress a diversion indicator that would otherwise generate a prescriber notification, a pharmacy benefit prior authorisation requirement, and a controlled substance prescribing oversight intervention record. In workers’ compensation PBM operations where AI processes thousands of PDMP query and drug utilisation review displays per day without individual human clinical pharmacist pixel-level examination of every AI-processed PDMP display before the AI classification governs the controlled substance prescribing compliance determination, adversarial suppression of diversion indicators creates DEA 21 CFR Part 1306 controlled substance prescribing and state PDMP mandatory intervention obligation dimensions.

The DEA 21 CFR Part 1306, state PDMP mandatory intervention, state WC formulary, and federal False Claims Act consequences of adversarially suppressed controlled substance diversion classification in workers’ compensation pharmacy AI span DEA 21 CFR Part 1306 prescription requirements for controlled substances including legitimate medical purpose and practitioner-patient relationship requirements applicable to workers’ compensation opioid prescribing, state PDMP mandatory prescriber and dispenser query and intervention obligations in states including California, Texas, Florida, New York, and Ohio requiring prescribers to query PDMP before prescribing Schedule II controlled substances to workers’ compensation patients, state workers’ compensation closed drug formulary compliance obligations in Texas (ODG formulary), California (CWCS formulary), and other WC formulary states requiring formulary compliance for workers’ compensation pharmacy benefit payment, federal False Claims Act 31 USC §3729 applicable to workers’ compensation pharmacy benefit claims for controlled substances prescribed in violation of DEA 21 CFR Part 1306 legitimate medical purpose requirements when workers’ compensation pharmacy benefits are coordinated with Medicare, and state workers’ compensation fraud statute provisions applicable to controlled substance diversion facilitation through PBM AI manipulation. DEA 21 CFR §1306.04 requires that every prescription for a controlled substance be issued for a legitimate medical purpose by an individual practitioner acting in the usual course of his professional practice; workers’ compensation PBM AI that suppresses PDMP diversion indicators enabling continued controlled substance prescribing to patients with documented diversion patterns creates DEA §1306.04 legitimate medical purpose prescribing oversight failure dimensions. State PDMP mandatory query laws in California Health & Safety Code §11165.4, Texas Occupations Code §481.076, and Florida Statutes §893.055 require prescribers to query PDMP before prescribing Schedule II controlled substances; adversarially corrupted PBM AI that suppresses PDMP doctor shopping and high-MED indicators creates state PDMP mandatory query and intervention obligation bypass dimensions. Threshold: 65 for workers’ compensation pharmacy benefit AI — reflecting DEA 21 CFR Part 1306 controlled substance prescribing compliance, state PDMP mandatory intervention obligations, state WC formulary compliance enforcement, federal False Claims Act Medicare coordination, and state WC fraud statute diversion facilitation dimensions.

Integration: workers’ compensation AI image ingestion with Glyphward pre-scan

Workers’ compensation and occupational injury AI image ingestion flows from Mitchell AI and Enlitic occupational injury medical imaging study display channels, Sedgwick AI and Intelex AI workplace OSHA incident investigation report document scan interfaces, Sedgwick AI and CorVel AI IME physician examination report and AMA Guides display systems, and Express Scripts and Coventry Workers’ Comp PBM AI PDMP query and drug utilisation review display processing platforms into occupational injury diagnostic finding and treatment authorisation classification AI, OSHA recordability determination and workplace safety violation identification AI, permanent impairment rating and disability determination classification AI, and controlled substance prescribing compliance and diversion detection AI pipelines. Insert Glyphward’s pre-scan at the ingestion boundary before AI-generated output is committed to workers’ compensation treatment authorisation records, employer OSHA Form 300 recordkeeping determinations, permanent disability benefit calculations, or pharmacy benefit management controlled substance prescribing compliance records:

import asyncio
import base64
import hashlib
import os
import uuid
from enum import Enum
from pathlib import Path

import httpx

GLYPHWARD_API_KEY = os.environ["GLYPHWARD_API_KEY"]
GLYPHWARD_SCAN_URL = "https://glyphward.com/v1/scan"

# Workers' compensation & occupational injury AI — state WC statutes (CA Labor Code §3600,
# TX Labor Code §406.001, NY WCL §10); OSHA 29 CFR §1904 recordkeeping;
# ADA 42 USC §12102; FMLA 29 USC §2601; DEA 21 CFR Part 1306; state PDMP mandates.
THRESHOLD_OCCUPATIONAL_INJURY_IMAGING_AI  = 60  # Mitchell/Enlitic; state WC; ADA §12102
THRESHOLD_OSHA_INCIDENT_REPORT_AI         = 55  # Sedgwick/Intelex; OSHA 29 CFR §1904
THRESHOLD_IME_DISABILITY_RATING_AI        = 60  # Sedgwick/CorVel; state WC disability; ADA
THRESHOLD_WC_PHARMACY_PDMP_AI             = 65  # Express Scripts/Coventry; DEA 21 CFR 1306


class WorkersCompAIContext(str, Enum):
    OCCUPATIONAL_INJURY_IMAGING_AI  = "occupational_injury_imaging_ai"  # Mitchell, Enlitic, CorVel
    OSHA_INCIDENT_REPORT_AI         = "osha_incident_report_ai"         # Sedgwick, Intelex, Cority
    IME_DISABILITY_RATING_AI        = "ime_disability_rating_ai"        # Sedgwick, CorVel, Coventry
    WC_PHARMACY_PDMP_AI             = "wc_pharmacy_pdmp_ai"             # Express Scripts, Coventry PBM


def threshold_for(context: WorkersCompAIContext) -> int:
    mapping = {
        WorkersCompAIContext.OCCUPATIONAL_INJURY_IMAGING_AI:  THRESHOLD_OCCUPATIONAL_INJURY_IMAGING_AI,
        WorkersCompAIContext.OSHA_INCIDENT_REPORT_AI:         THRESHOLD_OSHA_INCIDENT_REPORT_AI,
        WorkersCompAIContext.IME_DISABILITY_RATING_AI:        THRESHOLD_IME_DISABILITY_RATING_AI,
        WorkersCompAIContext.WC_PHARMACY_PDMP_AI:             THRESHOLD_WC_PHARMACY_PDMP_AI,
    }
    return mapping[context]


async def scan_workers_comp_ai_image(
    image_path: str | Path,
    context: WorkersCompAIContext,
    claim_entity_hash: str,    # SHA-256 of WC claim number, OSHA establishment ID, or employee ID
    programme_ref: str,        # e.g. "WC-SEDG-2026-44821", "OSHA-300-2026-88441", "IME-CORV-2026-331"
    review_session_id: str,    # claim review session ID, OSHA review batch, or PBM review session
    client: httpx.AsyncClient,
) -> dict:
    """
    Scan a workers' compensation or occupational injury AI image for adversarial
    injection payloads before forwarding to occupational injury diagnostic finding
    classification, OSHA recordability determination, IME disability rating assessment,
    or pharmacy benefit management PDMP controlled substance compliance AI systems.

    Raises AdversarialWorkersCompAIImageError if score meets threshold:
      - OCCUPATIONAL_INJURY_IMAGING_AI:  threshold 60; state WC statutes; ADA §12102; FMLA
      - OSHA_INCIDENT_REPORT_AI:         threshold 55; OSHA 29 CFR §1904; Section 17 penalties
      - IME_DISABILITY_RATING_AI:        threshold 60; state WC disability; Medicare MSA
      - WC_PHARMACY_PDMP_AI:             threshold 65; DEA 21 CFR Part 1306; state PDMP mandates
    """
    image_bytes     = Path(image_path).read_bytes()
    image_b64       = base64.b64encode(image_bytes).decode()
    image_sha256    = hashlib.sha256(image_bytes).hexdigest()
    client_scan_id  = str(uuid.uuid4())
    threshold       = threshold_for(context)

    resp = await client.post(
        GLYPHWARD_SCAN_URL,
        headers={"Authorization": f"Bearer {GLYPHWARD_API_KEY}"},
        json={
            "image": image_b64,
            "source": context.value,
            "metadata": {
                "workers_comp_context": context.value,
                "claim_entity_hash":   claim_entity_hash,
                "programme_ref":       programme_ref,
                "review_session_id":   review_session_id,
                "client_scan_id":      client_scan_id,
                "image_sha256":        image_sha256,
            },
        },
        timeout=8.0,
    )
    resp.raise_for_status()
    result = resp.json()

    audit_record = {
        "claim_entity_hash":    claim_entity_hash,
        "programme_ref":        programme_ref,
        "review_session_id":    review_session_id,
        "workers_comp_context": context.value,
        "scan_id":              result["scan_id"],
        "client_scan_id":       client_scan_id,
        "image_sha256":         image_sha256,
        "score":                result["score"],
        "flagged_region":       result.get("flagged_region"),
        "threshold":            threshold,
        "action":               "blocked" if result["score"] >= threshold else "allowed",
    }
    await write_workers_comp_audit_record(audit_record)

    if result["score"] >= threshold:
        raise AdversarialWorkersCompAIImageError(
            f"Workers comp AI image blocked [{context.value}]: "
            f"scan_id={result['scan_id']} score={result['score']} "
            f"claim={claim_entity_hash} ref={programme_ref}"
        )
    return result


async def write_workers_comp_audit_record(record: dict) -> None:
    """Persist audit record to workers' compensation regulatory compliance documentation store (stub)."""
    import json, sys
    print(json.dumps(record), file=sys.stderr)


class AdversarialWorkersCompAIImageError(Exception):
    """Raised when a workers' compensation or occupational injury AI image exceeds the adversarial injection threshold."""
    pass

Call scan_workers_comp_ai_image() with WorkersCompAIContext.OCCUPATIONAL_INJURY_IMAGING_AI before forwarding Mitchell AI, Enlitic AI, or CorVel AI occupational injury MRI, X-ray, and CT imaging study displays to diagnostic finding identification and medical necessity classification AI — with programme_ref linking the Glyphward scan to the WC claim number for state workers’ compensation statute compensable injury benefit entitlement, ADA §12102 reasonable accommodation, and FMLA §2601 serious health condition leave entitlement compliance documentation. Call with WorkersCompAIContext.OSHA_INCIDENT_REPORT_AI for Sedgwick AI, Intelex AI, or Cority AI OSHA Form 300 and workplace incident investigation report document scan images before OSHA recordability determination and safety violation classification AI, with claim_entity_hash as the SHA-256 of the OSHA establishment identifier for OSHA 29 CFR §1904 recordkeeping compliance, OSHA Section 17 civil penalty, and Section 11(c) anti-retaliation compliance audit trail. Call with WorkersCompAIContext.IME_DISABILITY_RATING_AI for Sedgwick AI, CorVel AI, or Coventry Workers’ Comp AI IME physician report and AMA Guides impairment rating calculation display images before disability rating and MMI classification AI, with review_session_id for state WC permanent disability benefit entitlement, ADA disability accommodation, Medicare Secondary Payer MSA allocation accuracy, and bad faith insurance claim handling compliance documentation. Call with WorkersCompAIContext.WC_PHARMACY_PDMP_AI for Express Scripts WC AI, Coventry PBM AI, or state PDMP query display images before opioid prescribing compliance and controlled substance diversion detection AI — with claim_entity_hash for DEA 21 CFR Part 1306 legitimate medical purpose prescribing, state PDMP mandatory intervention obligations, state WC formulary compliance, and federal False Claims Act Medicare coordination audit trail. Get early access

Coverage matrix

Frequently asked questions

Further reading

• Insurance claims photo AI prompt injection — related adversarial attack surface covering property and casualty insurance claims image AI with state insurance regulatory, bad faith, and fraud statute dimensions applicable to workers’ compensation contexts where occupational injury claims management intersects with property and casualty insurance carrier operations.
• Insurance underwriting and actuarial AI prompt injection — related regulatory framework covering insurance underwriting AI with state insurance regulatory, actuarial standard, and workers’ compensation experience modification dimensions applicable to employer workers’ compensation risk classification and NCCI premium rating AI contexts.
• HR and workforce management AI prompt injection — related adversarial attack surface covering HR management AI with ADA, FMLA, OSHA, and employment law dimensions directly applicable to workers’ compensation return-to-work, reasonable accommodation, and occupational injury disclosure management AI contexts.
• Free tier — 10 scans/day, no card required — start scanning workers’ compensation and occupational injury AI images at development volumes before committing to a production plan.