All Records

Harmful Content Generation at Scale: Fraudulent Services

Harmful Content Generation at Scale: Fraudulent Services is an AI risk in 4. Malicious Actors & Misuse focused on 4.3 > Fraud, scams, and targeted manipulati...

Inference Attacks

Inference Attacks is an AI risk in 2. Privacy & Security focused on 2.2 > AI system security vulnerabilities and attacks. It is most relevant during 2 - Post...

AI Agent Tool Invocation - AI Security Technique

AI Agent Tool Invocation is AI attack method AML.T0053 with evidence level: demonstrated. It includes 11 mitigations; 14 case studies.

Jailbreak of a model to subvert intended behavior

Jailbreak of a model to subvert intended behavior is an AI risk in 2. Privacy & Security focused on 2.2 > AI system security vulnerabilities and attacks. It...

Goal Hijacking

Goal Hijacking is an AI risk in 2. Privacy & Security focused on 2.2 > AI system security vulnerabilities and attacks. It is most relevant during 2 - Post-de...

AI Model Inference API Access - AI Security Technique

AI Model Inference API Access is AI attack method AML.T0040 with evidence level: realized. It includes 2 mitigations; 7 case studies; 19 related AI risks.

Risks from leaking or correctly inferring sensitive information

Risks from leaking or correctly inferring sensitive information is an AI risk in 2. Privacy & Security focused on 2.1 > Compromise of privacy by leaking or c...

LLM Prompt Crafting - AI Security Technique

LLM Prompt Crafting is AI attack method AML.T0065 with evidence level: realized. It includes 18 case studies.

Jailbreak in LLM Malicious Use - Backdoor Attack

Jailbreak in LLM Malicious Use - Backdoor Attack is an AI risk in 2. Privacy & Security focused on 2.2 > AI system security vulnerabilities and attacks. It i...

Exploiting Limited Generalization of Safety Finetuning

Exploiting Limited Generalization of Safety Finetuning is an AI risk in 2. Privacy & Security focused on 2.2 > AI system security vulnerabilities and attacks...

On Purpose - Pre-Deployment

On Purpose - Pre-Deployment is an AI risk in 2. Privacy & Security focused on 2.2 > AI system security vulnerabilities and attacks. It is most relevant durin...

Exfiltration via AI Agent Tool Invocation - AI Security Technique

Exfiltration via AI Agent Tool Invocation is AI attack method AML.T0086 with evidence level: realized. It includes 8 mitigations; 5 case studies; 26 related...

Jailbreak in LLM Malicious Use - Prompt Attacks

Jailbreak in LLM Malicious Use - Prompt Attacks is an AI risk in 2. Privacy & Security focused on 2.2 > AI system security vulnerabilities and attacks. It is...

Jailbreaking

Jailbreaking is an AI risk in 2. Privacy & Security focused on 2.2 > AI system security vulnerabilities and attacks. It is most relevant during 2 - Post-depl...

“Model Psychology” Attacks

“Model Psychology” Attacks is an AI risk in 2. Privacy & Security focused on 2.2 > AI system security vulnerabilities and attacks. It is most relevant during...

Adversarial AI: Circumvention of Technical Security Measures

Adversarial AI: Circumvention of Technical Security Measures is an AI risk in 2. Privacy & Security focused on 2.2 > AI system security vulnerabilities and a...

Prompt Leaking

Prompt Leaking is an AI risk in 2. Privacy & Security focused on 2.2 > AI system security vulnerabilities and attacks. It is most relevant during 2 - Post-de...

Evade AI Model - AI Security Technique

Evade AI Model is AI attack method AML.T0015 with evidence level: realized. It includes 6 mitigations; 17 case studies; 2 related AI risks.

Compromising privacy by correctly inferring private information

Compromising privacy by correctly inferring private information is an AI risk in 2. Privacy & Security focused on 2.1 > Compromise of privacy by leaking or c...

Data exfiltration

Data exfiltration is an AI risk in 2. Privacy & Security focused on 2.2 > AI system security vulnerabilities and attacks. It is most relevant during 2 - Post...