Prompt injection in veterinary and animal health AI

Four adversarial injection surfaces in veterinary and animal health AI

1. Veterinary radiology and digital X-ray AI injection (IDEXX AI Radiometry, VetCT AI, Vet-AI radiograph interpretation)

Veterinary radiology AI processes digital radiograph images, DICOM files converted to JPEG or PNG, and ultrasound image screenshots submitted through AI-assisted veterinary radiograph interpretation platforms to generate radiology reports that identify musculoskeletal injuries, thoracic and abdominal abnormalities, cardiac enlargement, pulmonary infiltrates, and oncological masses for small-animal practitioners. IDEXX Laboratories’ AI Radiometry platform — integrated into the IDEXX Digital Radiography (DR) system used at tens of thousands of veterinary practices — processes digital radiograph images captured through IDEXX Equiview and standard veterinary digital radiography detector panels, applying AI-assisted pattern recognition to generate structured radiology assessment reports that flag abnormalities including fractures, cardiomegaly, pneumothorax, pleural effusion, pulmonary infiltrates, and soft-tissue masses for veterinarian review. VetCT AI, a specialist veterinary teleradiology platform, processes DICOM files submitted by general practice veterinarians through a cloud-based teleradiology portal to board-certified veterinary radiologists who use AI-assisted triage and preliminary assessment tools to prioritise the case queue and generate specialist radiology reports. Vet-AI and similar AI radiograph interpretation services process digital radiograph image files submitted through API integrations with practice management systems including AVImark, ImproMed, and ClienTrax, providing AI-assisted radiograph interpretation at practices that do not have access to board-certified veterinary radiologists on staff or through a teleradiology service.

The digital radiograph image and DICOM file submission pathway is the adversarial injection surface: JPEG and PNG radiograph images exported from the practice’s digital radiography system and submitted through the IDEXX AI Radiometry cloud upload portal, the VetCT AI DICOM submission gateway, or the Vet-AI API integration point for AI-assisted radiology interpretation. An adversarially crafted digital radiograph image — in which pixel perturbations applied to the region of the radiograph showing a fracture callus, a soft-tissue mass with radiodense margins, an enlarged cardiac silhouette indicating cardiomegaly, or an increased pulmonary opacity pattern indicating pneumothorax — cause the IDEXX AI Radiometry or VetCT AI to generate a radiology report that does not flag the abnormality, characterising the film as within normal limits or attributing the relevant region to normal anatomical variation. The adversarial suppression motivation in veterinary radiology AI is clinical outcome driven: a radiology report that does not flag a neoplastic mass defers the oncological workup that would reveal the diagnosis, a report that misses a fracture defers the immobilisation and surgical intervention, and a report that misses cardiomegaly defers the cardiac evaluation and treatment that could prolong the animal’s life. The motivation for suppression may be a client who has already been told a diagnosis is likely and does not want it confirmed, a practice owner managing client relationships by avoiding difficult conversations, or a malicious actor with access to the image submission pathway at a high-volume veterinary practice.

Regulatory and professional consequences of adversarially suppressed veterinary radiology AI findings operate through AVMA and state veterinary board frameworks that impose affirmative diagnostic accuracy obligations on licensed veterinarians. The AVMA Guidelines for Veterinary Telemedicine (2021) require that veterinary telemedicine services — including AI-assisted radiograph interpretation services that constitute telemedicine under state veterinary practice act definitions — meet the same standard of care as in-person veterinary diagnosis, and that the veterinarian-client-patient relationship (VCPR) includes an obligation to provide accurate diagnostic assessment. State veterinary practice acts in all 50 states impose professional licensing obligations on veterinarians that include the duty to perform or obtain accurate diagnostic workups when a patient presents with signs or history consistent with a serious condition — a radiograph AI that suppresses a fracture or neoplastic finding may contribute to a diagnostic failure that constitutes veterinary malpractice under state law. Where the adversarial image manipulation is part of a pet insurance fraud scheme — suppressing the radiograph finding in the AI report to avoid a pre-existing condition exclusion determination — the manipulation may also constitute insurance fraud under state insurance fraud statutes. FTC Section 5 (15 USC § 45) deceptive practices authority extends to AI-assisted diagnostic services that misrepresent the completeness or accuracy of their diagnostic outputs. Threshold: 55 for veterinary radiology and digital X-ray AI (AVMA telemedicine guidelines, state VCPR requirements, AAVLD diagnostic standards, FTC Section 5, pet insurance fraud).

2. Veterinary pathology slide and cytology AI injection (IDEXX AI cytology, Antech Diagnostics AI, Zoetis AI pathology)

Veterinary pathology and cytology AI processes digitised histopathology slide images, cytology sample photograph scans, and blood smear image uploads submitted through AI-assisted veterinary diagnostic laboratory platforms to classify tissue and cellular samples as malignant, benign, inflammatory, infectious, or reactive — generating AI-assisted cytology and pathology reports that determine whether a veterinarian should proceed with oncological staging and treatment, additional diagnostic testing, or conservative monitoring. IDEXX’s AI cytology platform — integrated into the IDEXX Reference Laboratories network that serves over 65,000 veterinary practices — processes digitised cytology slide images from fine-needle aspirate (FNA) samples of lymph nodes, skin masses, splenic masses, and body cavity effusions, applying AI-assisted cell classification to generate preliminary cytology reports that assist board-certified veterinary clinical pathologists in their review and final diagnosis. Antech Diagnostics AI — operating over 2,500 veterinary reference laboratories globally under Mars Veterinary Health — processes histopathology slide scans and cytology photograph images for veterinary practices and specialty hospitals, using AI-assisted tissue classification to triage high-volume slide reading queues and generate AI-assisted preliminary diagnoses for pathologist review. Zoetis AI pathology tools process blood smear scans and cytology sample photographs for livestock diagnostic applications as well as companion animal diagnostics, integrating AI-assisted cell count and morphology classification with the Zoetis Diagnostics laboratory information management system used at veterinary reference laboratories globally.

The cytology slide scan and histopathology photograph submission pathway is the adversarial injection surface: digitised glass slide images and photograph scans of cytology preparations submitted through the IDEXX Reference Laboratories digital submission portal, the Antech Diagnostics AI laboratory information system, or the Zoetis Diagnostics API for AI-assisted pathology classification. An adversarially crafted cytology slide scan — in which pixel perturbations are applied to the cellular regions of the image that display the morphological hallmarks of malignancy — increased nuclear-to-cytoplasmic ratio, hyperchromatic nuclei with prominent nucleoli, anaplastic cell features including multinucleation and atypical mitotic figures, and irregular nuclear membranes — cause the IDEXX AI cytology or Antech Diagnostics AI to classify the sample as benign reactive lymphoid hyperplasia, benign reactive mesothelial cells, or a non-diagnostic sample requiring resubmission, when the actual cytology preparation demonstrates a high-grade lymphoma, carcinoma, or sarcoma requiring immediate oncological staging and treatment initiation. The adversarial suppression motivation in veterinary cytology AI can be financial — suppressing a malignancy finding to avoid the cost of oncology treatment — or may operate through a malicious actor at the laboratory submission pathway for a veterinary practice or boarding/breeding facility with repeated high-value patients.

AAVLD (American Association of Veterinary Laboratory Diagnosticians) laboratory accreditation standards impose diagnostic accuracy requirements on veterinary reference laboratories that process AI-assisted cytology and pathology reports, including requirements that AI-assisted preliminary assessments be reviewed by a board-certified veterinary clinical pathologist before the report is released to the submitting veterinarian. A cytology AI that is adversarially manipulated to produce a benign preliminary classification may cause the pathologist’s review to be perfunctory — because the preliminary AI classification anchors the pathologist’s assessment and the pathologist may not independently re-examine the digitised slide with the same scrutiny they would apply to a slide flagged by the AI as suspicious. The AVMA Principles of Veterinary Medical Ethics (Section IV, “Responsibilities to the Public and the Profession”) impose ethical obligations on veterinarians and veterinary service providers — including diagnostic laboratories — to provide diagnostic assessments that meet the applicable standard of care and to avoid misrepresentation of diagnostic findings. USDA APHIS diagnostic reporting requirements under 9 CFR Part 55 impose mandatory reporting obligations on veterinary diagnostic laboratories for certain zoonotic diseases — where the adversarially manipulated cytology AI suppresses findings consistent with a reportable zoonotic pathogen, the failure to report may constitute a federal regulatory violation. FDA CVM (Center for Veterinary Medicine) veterinary device guidance applicable to AI-assisted veterinary diagnostic tools addresses the accuracy and validation requirements for AI diagnostic devices in veterinary medicine. Threshold: 55 for veterinary pathology slide and cytology AI (AAVLD accreditation standards, AVMA ethics, state veterinary board malpractice, USDA APHIS zoonotic disease reporting, FDA CVM device guidance).

3. Livestock pharmaceutical batch record and compounding AI injection (Zoetis AI livestock, Merck Animal Health AI, Elanco Poultry AI, USDA FSIS AI)

Livestock pharmaceutical batch record and compounding AI processes pharmaceutical compounding record photographs, drug batch record document scans, and veterinary drug label photographs submitted through AI-assisted livestock pharmaceutical management and USDA food safety compliance platforms to extract active pharmaceutical ingredient (API) concentrations, verify batch specification compliance, and generate pharmaceutical use records for livestock operations subject to USDA FSIS Residue Avoidance Program requirements. Zoetis — whose livestock pharmaceutical portfolio spans swine, poultry, cattle, and aquaculture and includes major products such as Draxxin (tulathromycin), Excede (ceftiofur), and Advocin (danofloxacin) — operates AI-assisted pharmaceutical management tools for livestock producers that process compounding batch record photographs and drug use documentation through AI-assisted compliance verification platforms integrated with veterinary prescription management systems. Merck Animal Health AI — whose livestock portfolio includes Nuflor (florfenicol), Baytril (enrofloxacin), and Safeguard (fenbendazole) — processes pharmaceutical batch record documents and drug concentration verification photographs through AI-assisted livestock pharmaceutical compliance tools for large integrated livestock producers. Elanco Animal Health AI — with a significant poultry and swine pharmaceutical portfolio including Tylan (tylosin) and Corid (amprolium) — processes drug batch records and compounding documentation for AI-assisted pharmaceutical compliance management in poultry integrators and contract swine production operations.

The pharmaceutical compounding batch record photograph and drug label scan submission pathway is the adversarial injection surface: photographs of paper compounding batch records, drug concentration verification printouts, and veterinary drug label images submitted through Zoetis AI, Merck Animal Health AI, or Elanco AI pharmaceutical compliance platforms for AI-assisted API concentration extraction and specification compliance verification. An adversarially crafted compounding batch record photograph — in which pixel perturbations applied to the printed drug concentration field — such as the active oxytetracycline concentration in milligrams per litre in a medicated water system record, the injectable tylosin concentration per millilitre in a compounded product, or the withdrawal time calculation based on the drug concentration and species body weight — cause the Zoetis AI or Merck Animal Health AI to extract a within-specification API concentration from the batch record image when the actual batch record documents an out-of-specification compounding event that results in a drug residue violation risk in animals destined for slaughter. The adversarial manipulation motivation in livestock pharmaceutical compliance AI is food safety regulatory driven: USDA FSIS residue violations — where tissue samples from slaughtered animals contain drug residues above the established safe residue tolerance — generate FSIS Notice detentions, facility holds, carcass condemnation, and in chronic cases, loss of USDA inspection approval that can shut down a livestock production operation.

USDA FSIS regulatory consequences for adulterated livestock are among the most serious in agricultural food safety law. The USDA FSIS National Residue Program (9 CFR Part 310, FSIS Notice 12-25) conducts residue testing on harvested livestock carcasses, and animals found to contain violative drug residues above safe tolerance levels established under 21 CFR Part 556 are condemned and the producing operation is placed on a FSIS Residue Repeat Violator List — generating increased testing frequency that creates operational disruption and reputational consequences for the livestock producer. The AMDUCA (Animal Medicinal Drug Use Clarification Act of 1994, 21 USC § 360b note) and its implementing regulations at 21 CFR Part 530 govern extralabel drug use in food-producing animals and impose strict withdrawal time requirements for drugs used in livestock, with violations constituting adulteration of food under the Federal Food, Drug, and Cosmetic Act (21 USC § 342). USDA FSIS criminal prosecution for adulterated meat under the Federal Meat Inspection Act (21 USC § 621) imposes criminal penalties for the distribution of adulterated meat products, including penalties of up to 3 years imprisonment and $10,000 fine per violation for knowing introduction of adulterated meat into commerce. An adversarially manipulated livestock pharmaceutical compliance AI that generates a within-specification batch record extraction from a manipulated out-of-specification batch record photograph — causing the livestock producer to maintain animals in production through a withdrawal period that is too short for the actual drug concentration — creates a regulatory and criminal exposure pathway under USDA FSIS, AMDUCA, and the FMIA that is not eliminated by the AI mediation of the false documentation. Threshold: 55 for livestock pharmaceutical batch record and compounding AI (USDA FSIS Residue Avoidance Program, AMDUCA 21 CFR Part 530, FDA CVM NADA/ANADA, 9 CFR Part 310, FMIA 21 USC § 621).

4. Pet insurance claim photograph AI injection (Trupanion AI, Nationwide Pet Insurance AI, ASPCA Pet Health Insurance AI, Lemonade Pet AI)

Pet insurance claim photograph AI processes injury and condition photograph submissions, veterinary invoice photographs, and medical record page photographs submitted by policyholders through AI-assisted pet insurance claim assessment platforms to evaluate claim severity, verify treatment necessity, classify injury or disease type, and generate reimbursement tier determinations that are used in automated claim approval workflows. Trupanion — the largest publicly traded pet insurance provider in North America with over 950,000 enrolled pets and approximately $1 billion in annualised premium — operates AI-assisted claim processing tools that evaluate submitted veterinary invoice photographs and medical record images to classify claims by condition category and severity and generate AI-assisted eligibility and reimbursement recommendations integrated into its Express Claims system, which processes a significant percentage of eligible claims in real time at the point of payment at participating veterinary practices. Nationwide Pet Insurance — the largest US pet insurer by enrolled pet count with over 1 million policies — processes injury and condition photograph submissions through AI-assisted claim triage and assessment tools that classify submitted images by injury type and severity to route claims to automated approval, adjuster review, or veterinary specialist review queues. ASPCA Pet Health Insurance (underwritten by PTZ Insurance Agency) and Lemonade Pet Insurance — both major direct-to-consumer pet insurance providers with growing AI-assisted claim processing capabilities — process policyholder-submitted injury photographs and veterinary document scans through AI-assisted claim assessment tools integrated with their mobile claim submission workflows.

The pet insurance claim photograph submission pathway is the adversarial injection surface: injury condition photographs, surgical site images, and veterinary invoice photographs submitted by policyholders through the Trupanion Express Claims portal, the Nationwide Pet Insurance mobile claim app, the ASPCA Pet Health Insurance claim submission interface, or the Lemonade Pet AI claim chatbot for AI-assisted claim severity assessment and reimbursement tier classification. An adversarially crafted injury or surgical site photograph — in which pixel perturbations applied to the wound margin region, the tissue visualisation field, or the surgical incision documentation image cause the Trupanion AI or Nationwide Pet Insurance AI to classify a minor laceration requiring basic wound closure as a severe traumatic injury requiring extensive surgical repair, or to classify a routine dental cleaning procedure photograph as a more complex oral surgery — generates an inflated claim estimate that exceeds the reimbursement that a human claim adjuster with the original unmanipulated image would authorise. The adversarial claim inflation motivation is financial fraud: pet insurance claim fraud through image manipulation is an extension of the broader insurance fraud pattern of inflating claim severity through document manipulation — the AI mediation of the inflated claim merely replaces the human adjuster who would have been the target of the manipulated submission in a pre-AI insurance workflow.

State insurance fraud statutory consequences for pet insurance claim manipulation are substantive and broadly applicable: pet insurance fraud is prosecutable under general insurance fraud statutes in all 50 states, which criminalise the knowing submission of false or fraudulent claims or supporting documentation to an insurance company for the purpose of obtaining benefits to which the claimant is not entitled. The NAIC (National Association of Insurance Commissioners) pet insurance model regulation — adopted in more than 20 states as of 2025 — establishes standardised definitions for covered conditions and claim documentation requirements that impose additional legal structure on the pet insurance claim submission process. California DOI Fair Claims Settlement Practices regulations (10 CCR § 2695) impose specific requirements on insurers and by extension inform the claim documentation standards applicable to California policyholders. FTC Section 5 (15 USC § 45) deceptive practices authority extends to AI-assisted claim processing that misrepresents the accuracy or completeness of claim severity assessments — both from the insurer’s perspective (unfair AI denials) and potentially from a fraud investigation perspective where AI claim inflation tools are marketed or facilitated. A pet insurer that implements Glyphward pre-scan for claim photograph inputs to its Trupanion/Nationwide/Lemonade AI claim platform has a documented basis for demonstrating that it implemented a data integrity verification measure in its AI claim processing workflow — relevant to regulatory examinations by state insurance departments under the NAIC model regulation and to any insurer defence in a bad-faith insurance litigation proceeding where AI claim assessment accuracy is challenged. Threshold: 55 for pet insurance claim photograph AI (state insurance fraud statutes in all 50 states, NAIC pet insurance model regulation, California DOI 10 CCR § 2695, FTC Section 5 deceptive AI claim processing).

Integration: veterinary and animal health AI image ingestion with Glyphward pre-scan

Veterinary and animal health AI image ingestion flows from IDEXX AI Radiometry and VetCT AI DICOM upload portals, Antech Diagnostics and IDEXX Reference Laboratories cytology slide scan submission interfaces, Zoetis/Merck/Elanco livestock pharmaceutical batch record photograph APIs, and Trupanion/Nationwide/Lemonade pet insurance claim photograph submission portals into AI radiology interpretation, cytology classification, pharmaceutical compliance, and insurance claim assessment pipelines. Insert Glyphward’s pre-scan at the ingestion boundary — in all veterinary and animal health AI contexts, where the patient welfare, food safety, and insurance fraud consequences of adversarial image manipulation are categorically significant:

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"

# Veterinary and animal health AI — AVMA telemedicine guidelines,
# AAVLD diagnostic accreditation standards, USDA FSIS Residue Avoidance
# Program, AMDUCA 21 CFR Part 530, FDA CVM device guidance, state
# insurance fraud statutes, NAIC pet insurance model regulation.
# Threshold 55 — patient welfare, food safety (livestock), and insurance
# fraud consequences of adversarial false negatives exceed operational
# cost of human review of borderline flagged images.
THRESHOLD_VET = 55


class VeterinaryAIContext(str, Enum):
    RADIOLOGY            = "radiology"            # IDEXX AI Radiometry, VetCT AI, Vet-AI
    CYTOLOGY_PATHOLOGY   = "cytology_pathology"   # IDEXX AI cytology, Antech Diagnostics AI, Zoetis AI
    LIVESTOCK_PHARMA     = "livestock_pharma"     # Zoetis AI livestock, Merck Animal Health AI, Elanco Poultry AI
    PET_INSURANCE_CLAIM  = "pet_insurance_claim"  # Trupanion AI, Nationwide Pet, ASPCA, Lemonade Pet AI


async def scan_veterinary_image(
    image_path: str | Path,
    context: VeterinaryAIContext,
    practice_id_hash: str,      # SHA-256 of AVID practice ID or IDEXX account number
    patient_hash: str,          # SHA-256 of patient record ID — never raw patient data
    case_ref: str,              # e.g. "radiol_canine_thorax_2026Q2", "cyto_FNA_lymph_001"
    client: httpx.AsyncClient,
) -> dict:
    """
    Scan a veterinary or animal health AI image for adversarial injection
    payloads before forwarding to veterinary radiology interpretation AI,
    cytology and pathology classification AI, livestock pharmaceutical
    compliance AI, or pet insurance claim assessment AI.

    Raises AdversarialVeterinaryImageError if the Glyphward score meets or
    exceeds the veterinary threshold (55).
    """
    image_bytes = Path(image_path).read_bytes()
    image_b64 = base64.b64encode(image_bytes).decode()
    image_sha256 = hashlib.sha256(image_bytes).hexdigest()
    scan_id = str(uuid.uuid4())

    resp = await client.post(
        GLYPHWARD_SCAN_URL,
        headers={"Authorization": f"Bearer {GLYPHWARD_API_KEY}"},
        json={
            "image": image_b64,
            "source": context.value,
            "metadata": {
                "vet_context": context.value,
                "practice_id_hash": practice_id_hash,
                "patient_hash": patient_hash,
                "case_ref": case_ref,
                "client_scan_id": scan_id,
                "image_sha256": image_sha256,
            },
        },
        timeout=10.0,
    )
    resp.raise_for_status()
    result = resp.json()

    audit_record = {
        "practice_id_hash": practice_id_hash,
        "patient_hash": patient_hash,
        "case_ref": case_ref,
        "vet_context": context.value,
        "scan_id": result["scan_id"],
        "client_scan_id": scan_id,
        "image_sha256": image_sha256,
        "score": result["score"],
        "flagged_region": result.get("flagged_region"),
        "threshold": THRESHOLD_VET,
        "action": "blocked" if result["score"] >= THRESHOLD_VET else "allowed",
    }
    await write_vet_compliance_record(audit_record)

    if result["score"] >= THRESHOLD_VET:
        raise AdversarialVeterinaryImageError(
            f"Veterinary AI image blocked [{context.value}]: "
            f"scan_id={result['scan_id']} score={result['score']} "
            f"practice_hash={practice_id_hash} ref={case_ref}"
        )
    return result


async def scan_cytology_batch(
    slide_paths: list[Path],
    practice_id_hash: str,
    patient_hash: str,
    case_ref: str,
) -> dict:
    """
    Scan all digitised cytology or histopathology slide images for a
    diagnostic case before loading into IDEXX AI cytology or Antech
    Diagnostics AI pathology classification. All slides scanned with
    CYTOLOGY_PATHOLOGY context (threshold 55).
    """
    allowed, blocked, errors = [], [], []
    async with httpx.AsyncClient() as client:
        tasks = [
            scan_veterinary_image(
                p, VeterinaryAIContext.CYTOLOGY_PATHOLOGY,
                practice_id_hash, patient_hash,
                f"{case_ref}_slide{i:04d}", client,
            )
            for i, p in enumerate(slide_paths)
        ]
        results = await asyncio.gather(*tasks, return_exceptions=True)

    for path, result in zip(slide_paths, results):
        if isinstance(result, AdversarialVeterinaryImageError):
            blocked.append({"path": str(path), "error": str(result)})
        elif isinstance(result, Exception):
            errors.append({"path": str(path), "error": str(result)})
        else:
            allowed.append({"path": str(path), "scan_id": result["scan_id"]})

    return {
        "practice_id_hash": practice_id_hash,
        "patient_hash": patient_hash,
        "case_ref": case_ref,
        "total": len(slide_paths),
        "allowed": len(allowed),
        "blocked": len(blocked),
        "errors": len(errors),
        "blocked_slides": blocked,
    }


async def write_vet_compliance_record(record: dict) -> None:
    """Persist compliance audit record to practice/laboratory records system (stub)."""
    import json, sys
    print(json.dumps(record), file=sys.stderr)


class AdversarialVeterinaryImageError(Exception):
    """Raised when a veterinary AI image exceeds the adversarial injection threshold."""
    pass

Call scan_veterinary_image() with VeterinaryAIContext.RADIOLOGY for digital radiograph JPEG/PNG exports and DICOM-derived image files before IDEXX AI Radiometry, VetCT AI, or Vet-AI radiograph interpretation — this is the highest patient welfare integration point in the small-animal diagnostic AI pipeline because an adversarially suppressed fracture or neoplastic mass finding defers the clinical intervention that determines the animal’s outcome. Call scan_cytology_batch() for digitised cytology and histopathology slide image sets before IDEXX Reference Laboratories AI cytology or Antech Diagnostics AI pathology classification — batch slide scanning prevents adversarial suppression of the malignancy morphological markers that would cause the AI preliminary classification to anchor the pathologist’s review toward a benign misdiagnosis. Call scan_veterinary_image() with VeterinaryAIContext.LIVESTOCK_PHARMA for all pharmaceutical compounding batch record photographs and drug label scans before Zoetis AI, Merck Animal Health AI, or Elanco AI pharmaceutical compliance extraction — livestock pharmaceutical compliance document scanning has direct USDA FSIS and AMDUCA exposure because the AI-extracted batch record values become the regulatory compliance record used to determine withdrawal time compliance for food-safety livestock. Call with VeterinaryAIContext.PET_INSURANCE_CLAIM for all injury, surgical site, and medical record page photographs submitted through Trupanion, Nationwide, ASPCA, or Lemonade Pet AI claim portals before AI severity classification — pet insurance claim photograph scanning prevents adversarial severity inflation that constitutes insurance fraud under state statutes in all 50 states. The Glyphward audit record should be retained as part of the veterinary practice’s or insurer’s compliance records for the applicable regulatory retention period. Get early access

Coverage matrix

Frequently asked questions

Further reading

• Indirect prompt injection via image — foundational attack pattern underlying all four veterinary AI injection surfaces; covers how adversarial pixel-level perturbations cause AI misclassification through image content manipulation without detectable visual artefacts visible to image quality screening.
• Prompt injection in healthcare AI — human healthcare AI with closely related diagnostic image injection vectors in radiology and pathology AI; covers HIPAA, FDA SaMD, and medical device adversarial image injection frameworks that parallel AVMA and AAVLD veterinary standards.
• Prompt injection in clinical trials and pharmaceutical AI — pharmaceutical and clinical research AI with overlapping drug batch record document injection vectors relevant to AMDUCA and FDA CVM livestock pharmaceutical compliance AI contexts.
• Prompt injection scanner for document AI — document AI scanning covering the broader class of scanned compliance record and laboratory result document injection vectors applicable to livestock pharmaceutical batch record and pet insurance claim document submission AI.
• Free tier — 10 scans/day, no card required — start scanning veterinary AI images at development volumes before committing to a production plan.