Misinformation & Hallucination
False or fabricated outputs that can mislead users or the public.
- Fake citations
- Fabricated facts
- Synthetic news
Risk Categories
SAIFE’s working taxonomy. These categories anchor detection, dashboards, and simulators.
Misinformation & Hallucination
False or fabricated outputs that can mislead users or the public.
Examples: Fake citations · Fabricated facts · Synthetic news
Privacy & Data Leakage
Exposing personal or sensitive data in prompts, outputs, or logs.
Examples: PII leakage · Training data exposure · Prompt injection exfiltration
Bias & Discrimination
Unfair treatment or stereotyping across protected attributes.
Examples: Toxic outputs · Unequal performance · Stereotyped recommendations
Safety & Harmful Content
Outputs that enable or encourage harm to self or others.
Examples: Violence instructions · Self-harm content · Illegal activity guidance
Security & Abuse
Model/system exploits, jailbreaking, or unauthorized access.
Examples: Prompt injection · RCE chains · Account takeover patterns
IP & Copyright
Infringing content or non-transformative reproduction.
Examples: Verbatim outputs · Style cloning · Trademark misuse
Fraud & Impersonation
Deceptive identity or financial abuse via AI systems.
Examples: Deepfake scams · Phishing text · Synthetic voice spoofing
Child Safety
Any material or behavior that endangers minors.
Examples: Sexualized content · Grooming patterns · Exploitative imagery
Medical & Legal Advice
Unqualified guidance in high-stakes domains.
Examples: Unsafe dosing advice · Unauthorized legal counsel
Political Manipulation
Coordinated influence or civic process interference.
Examples: Astroturfing · Targeted persuasion · Voter suppression content
Privacy & Data Leakage
Exposing personal or sensitive data in prompts, outputs, or logs.
- PII leakage
- Training data exposure
- Prompt injection exfiltration
Bias & Discrimination
Unfair treatment or stereotyping across protected attributes.
- Toxic outputs
- Unequal performance
- Stereotyped recommendations
Safety & Harmful Content
Outputs that enable or encourage harm to self or others.
- Violence instructions
- Self-harm content
- Illegal activity guidance
Security & Abuse
Model/system exploits, jailbreaking, or unauthorized access.
- Prompt injection
- RCE chains
- Account takeover patterns
IP & Copyright
Infringing content or non-transformative reproduction.
- Verbatim outputs
- Style cloning
- Trademark misuse
Fraud & Impersonation
Deceptive identity or financial abuse via AI systems.
- Deepfake scams
- Phishing text
- Synthetic voice spoofing
Child Safety
Any material or behavior that endangers minors.
- Sexualized content
- Grooming patterns
- Exploitative imagery
Medical & Legal Advice
Unqualified guidance in high-stakes domains.
- Unsafe dosing advice
- Unauthorized legal counsel
Political Manipulation
Coordinated influence or civic process interference.
- Astroturfing
- Targeted persuasion
- Voter suppression content