All threats

ai-model-extraction

Attacker queries AI model systematically to reconstruct or steal proprietary model weights and architecture

MITRE ATT&CK techniques

Common Weakness Enumeration

• CWE-200

Mitigating controls

ctrl-extract-1

Implement strict rate limiting on model API endpoints

ctrl-extract-2

Monitor for systematic querying patterns

ctrl-extract-3

Add output perturbation to prevent exact replication

ctrl-extract-4

Restrict API access with strong authentication

References

• https://genai.owasp.org/llm-top-10/
• https://atlas.mitre.org/