AI Security Taxonomy

Threat Models

• Threat Models
• Prompt injection and tool abuse
• Data poisoning and supply-chain compromise
• Model extraction and inversion
• Privacy leakage
• Jailbreaks and policy bypass
• Agent autonomy and delegation risk

Evidence Types

• Peer-reviewed paper
• Preprint
• Benchmark report
• Incident report
• Official advisory or standard
• Reproducible code repository
• Expert commentary

System Layers

• Data
• Training
• Model weights
• Inference API
• Prompt and context
• Tool interface
• Agent memory
• Deployment and monitoring
• Governance and response