AI SECURITY CASES × FRAMEWORKS
Cisco AI Security & Safety Framework v2.0 · Categories and AI Security Cases
Explore AI security cases, sources, and defensive analysis across 19 populated Cisco AI Security & Safety Framework v2.0 categories.
Cisco v2.0.0 organizes AI security and safety risks into 19 objectives, 41 techniques, and 115 subtechniques. Objective summaries below are written by AIDEFEND from the official child techniques; identifiers and names follow Cisco. Listed controls and counts include only actionable AIDEFEND defenses assessed as Full or High relevance.
- 1 · Goal Hijacking (Prompt Injection, Jailbreak) 86 cases
- 2 · Agentic Autonomy Failures (Misuse, Misalignment, Drift) 15 cases
- 3 · Masquerading / Obfuscation / Impersonation 3 cases
- 4 · Communication Compromise 9 cases
- 5 · Persistence 27 cases
- 6 · Feedback Loop Manipulation 8 cases
- 7 · Sabotage / Integrity Degradation 12 cases
- 8 · Data and Privacy Violations 99 cases
- 9 · Supply Chain Compromise 54 cases
- 10 · Model Theft / Extraction 9 cases
- 11 · Adversarial Evasion 31 cases
- 12 · Action-Space and Integration Abuse 105 cases
- 13 · Availability Abuse 9 cases
- 14 · Privilege Compromise 101 cases
- 15 · Harmful / Misleading / Inaccurate Content 8 cases
- 16 · Surveillance 3 cases
- 17 · Cyber-Physical / Sensor Attacks 4 cases
- 18 · System Misuse / Malicious Application 21 cases
- 19 · Multimodal / Cross-Modal Risks 2 cases
An empty category does not establish safety. A category relation does not prove a defense is deployed or effective.