Threat Models

Threat models define attacker capability, target system boundary, objective, and success criteria.

Template

• Actor:
• Access:
• Knowledge:
• Control:
• Objective:
• Success criteria:
• Observable evidence:
• Relevant mitigations:

Initial Families

• External user attacks against hosted models
• Indirect prompt injection through retrieved or browsed content
• Insider or supply-chain compromise during training and fine-tuning
• Model provider extraction or privacy attacks by API users
• Autonomous agent tool abuse or unsafe delegation