SafetyCategory
Enums · AiDotNet.Enums
Comprehensive taxonomy of safety and harm categories for content classification.
For Beginners
These are the specific types of harmful content that the safety system can detect. Each category represents a different kind of risk, from hate speech to deepfakes to copyright violations. You can configure which categories to check for and what action to take for each.
How It Works
This enum defines a hierarchical taxonomy of content safety categories covering all known types of harmful, inappropriate, or policy-violating content across text, image, audio, and video modalities.
References:
- UnsafeBench 11-category taxonomy (Qu et al., 2024)
- WildGuard 13-risk-category classification (Allen AI, 2024)
- OmniSafeBench-MM 9 risk domains with 50 fine-grained categories (2025)
- MM-SafetyBench 13 scenarios (Liu et al., ECCV 2024)
- EU AI Act risk classification (Articles 5, 6, 50, 52)
Fields
| Field | Summary |
|---|---|
SafetyCategory AIGenerated |
Content detected as AI-generated (text, image, audio, or video). |
SafetyCategory Bias |
Biased content that treats demographic groups inequitably. |
SafetyCategory CopyrightViolation |
Content that infringes copyright or reproduces protected works. |
SafetyCategory Deepfake |
Deepfake content — AI-generated or manipulated media impersonating real people. |
SafetyCategory Dehumanization |
Content that dehumanizes or degrades individuals or groups. |
SafetyCategory Discrimination |
Discriminatory content or policies targeting protected groups. |
SafetyCategory Disinformation |
Disinformation — deliberately false content intended to deceive. |
SafetyCategory Doxxing |
Doxxing — publishing private information to identify or locate someone. |
SafetyCategory DrugManufacturing |
Instructions for manufacturing illegal drugs. |
SafetyCategory FinancialAdvice |
Unqualified financial advice that could cause harm. |
SafetyCategory Fraud |
Fraud, scam, or social engineering content. |
SafetyCategory Hallucination |
Hallucinated or fabricated content not grounded in facts or source material. |
SafetyCategory Harassment |
Targeted harassment or bullying of individuals. |
SafetyCategory HateSpeech |
Hate speech targeting protected groups based on race, religion, gender, etc. |
SafetyCategory IllegalActivities |
Instructions or promotion of illegal activities. |
SafetyCategory Impersonation |
Impersonation of real individuals or organizations. |
SafetyCategory JailbreakAttempt |
Jailbreak attempt — attempts to bypass safety measures. |
SafetyCategory LegalAdvice |
Unqualified legal advice that could cause harm. |
SafetyCategory Malware |
Malware, hacking tools, or cyberattack instructions. |
SafetyCategory Manipulated |
Content that has been digitally manipulated or altered. |
SafetyCategory MedicalAdvice |
Unqualified medical advice that could cause harm. |
SafetyCategory Misinformation |
Misinformation — unintentionally false or misleading content. |
SafetyCategory ModelExtraction |
Attempts to extract or replicate the model itself. |
SafetyCategory PIIExposure |
Exposure of personally identifiable information (PII). |
SafetyCategory PolicyViolation |
Content that violates a topic restriction or custom policy rule. |
SafetyCategory PromptInjection |
Prompt injection — attempts to override system instructions. |
SafetyCategory SexualExplicit |
Sexually explicit content including pornography or graphic sexual acts. |
SafetyCategory SexualMinors |
Child sexual abuse material (CSAM). |
SafetyCategory SexualSuggestive |
Sexually suggestive content that is not explicitly graphic. |
SafetyCategory SocialEngineering |
Social engineering content designed to manipulate victims. |
SafetyCategory Stereotyping |
Stereotyping content that reinforces harmful generalizations. |
SafetyCategory SurveillanceEnabling |
Content that enables mass surveillance. |
SafetyCategory TrainingDataLeakage |
Leakage of memorized training data. |
SafetyCategory TransparencyViolation |
Content that lacks transparency about AI involvement (regulatory compliance). |
SafetyCategory ViolenceGraphic |
Graphic depictions of violence, gore, or injury. |
SafetyCategory ViolenceSelfHarm |
Content promoting, glorifying, or instructing self-harm. |
SafetyCategory ViolenceSuicide |
Content promoting, glorifying, or instructing suicide. |
SafetyCategory ViolenceTerrorism |
Content promoting, glorifying, or instructing terrorism. |
SafetyCategory ViolenceThreat |
Threats of violence against individuals or groups. |
SafetyCategory ViolenceWeapons |
Content promoting or depicting weapons use. |
SafetyCategory Watermarked |
Content that contains a digital watermark (informational, not harmful). |
SafetyCategory WeaponsInstructions |
Instructions for manufacturing or using weapons. |
int value__ |