AI Reality Check verifies AI-generated content using live web search to find real evidence for each claim. Results are then classified against four internationally recognized risk frameworks.
National Institute of Standards and Technology
The NIST AI Risk Management Framework provides a structured approach to managing AI risks. We evaluate claims against NIST trustworthiness characteristics: validity & reliability, safety, security & resilience, accountability & transparency, explainability & interpretability, privacy enhancement, and fairness.
View official document →European Parliament and Council
The EU AI Act establishes a risk-based regulatory framework for AI systems. We classify each claim into one of four risk levels — minimal, limited, high, or unacceptable — based on the potential harm if the claim is false or misleading. Claims about health, safety, finance, and legal matters receive higher risk classifications.
View official document →Open Worldwide Application Security Project
OWASP identifies the most critical security risks for LLM applications. We use this framework to detect hallucination patterns, training data contamination signals, prompt injection artifacts, and output fabrication indicators in AI-generated text.
View official document →International Organization for Standardization
ISO/IEC 42001 specifies requirements for an AI management system. We reference this standard for responsible AI governance practices, quality assurance of AI outputs, and establishing trust through systematic management of AI-related risks.
View official document →Claim Decomposition: The submitted text is broken down into individual, verifiable factual claims.
Web Search Verification: Each claim is searched on the live web for supporting or contradicting evidence. Real sources are collected and used to assess whether the claim is verified, contradicted, or unverifiable.
Risk Classification: Each claim is classified under the EU AI Act risk framework (minimal, limited, high, unacceptable) based on its potential for harm if false.
Source Citation: Real URLs from web search results are provided for each claim, linking directly to the evidence used for verification.
Weighted Scoring: The overall trust score (0-100) is calculated with severity weighting — false claims in high-risk categories (health, safety, finance) are penalized more heavily than inaccurate trivia.