ArtedEye — agentic threat model

Not certain from the listing — likely uses specialized computer vision or latent diffusion models for image enhancement. Threats include adversarial inputs (e.g., crafted images designed to exploit model vulnerabilities) and model stealing of proprietary enhancement weights.

Processes highly sensitive biometric data (eye/iris images). Threats include data exfiltration of raw or enhanced biometric templates, unauthorized retention of user images, and failure to properly isolate user data in the processing pipeline.

Not certain from the listing — likely uses a simple sequential pipeline rather than a complex agentic framework. Threats include insecure integration with image processing libraries and potential buffer overflows or remote code execution via malformed image uploads.

Not certain from the listing — likely hosted on cloud infrastructure with GPU acceleration. Threats include container compromise, unauthorized access to cloud storage buckets containing user eye photos, and insecure API endpoints for the upload/download flow.

The agent utilizes a human-in-the-loop (HITL) manual review process for quality assurance. However, it is unclear if there are automated guardrails to detect and block non-eye images, explicit content, or adversarial payloads before they reach the model or human reviewers.

The listing explicitly claims GDPR compliance. Because iris/eye patterns are classified as biometric data under GDPR Article 9, the agent must enforce strict consent, data minimization, and right-to-be-forgotten mechanisms. Threats include compliance violations and lack of audit logging for biometric access.

Not certain from the listing — appears to operate as a standalone vertical application with no multi-agent or marketplace interactions. The primary ecosystem threat is limited to secure integration with third-party payment processors for unlocking downloads.