AI & LLM Threat Intelligence
A dedicated mapping of AI-centric threats combining the MITRE ATLAS framework and the official OWASP Top 10 for LLM Applications (2025).
Reconnaissance
Gathering information about the AI system
Resource Development
Setting up resources to target the AI system
Initial Access
Gaining access to the AI system or environment
ML Model Access
Gaining read or write access to the ML model
Execution
Running malicious code or inputs within the AI environment
Persistence
Maintaining access to the AI system across restarts
Privilege Escalation
Gaining higher-level permissions in the AI environment
Defense Evasion
Avoiding detection while attacking the AI system
Credential Access
Stealing credentials used by the AI system
Discovery
Learning the internal structure of the AI system
Collection
Gathering data, model weights, or artifacts
ML Attack Staging
Staging adversarial inputs or poisoned data
Exfiltration
Stealing the model, datasets, or intel
Impact
Degrading, manipulating, or denying the AI service