AI Threat Model

Test against:

• Prompt injection
• Malicious strategy input
• Manipulated market data
• AI hallucination
• Invalid intents
• Replay
• Privilege escalation
• Backend compromise
• Provider compromise

The deterministic security layers must remain effective even if the AI layer is compromised.