Defacta
Defacta is an independent verification layer that checks AI-generated or external text against credible sources before it informs decisions, publications, or filings. It replicates the human verification process as a structured, multi-stage pipeline, using LLMs only to accelerate specific steps while constraining them with fresh data retrieval and source-grounded analysis.
The platform extracts key claims from pasted text, articles, URLs, or AI outputs and checks them against independent sources, fact-checking registries, and credibility signals. It flags hallucinations, fabricated citations, unsupported assertions, bias, manipulation patterns, and misleading framing.
Reports combine factual findings, source metadata, risk indicators, and recommended actions. Every report is timestamped with SHA-256 hashes, creating a durable audit trail. Users can edit text, generate repair prompts, re-check changes, and compare versions.
Learn more
Facticity.AI
Facticity.AI is an AI-powered fact-checking platform that verifies claims, detects misinformation, and delivers evidence-backed insights in real time. Users can submit text or video links from YouTube, Instagram, and TikTok to receive verdicts with cited sources, bias context, and clear explanations. Built for a multilingual, multimodal world, Facticity.AI processes written and video content across multiple languages, with particular strength in Asian languages. Sources are surfaced across supporting, opposing, and neutral perspectives, and cross-referenced against political bias signals for balanced context. Benchmarked at 98.3% accuracy, Facticity.AI outperforms leading AI search tools and was named TIME's Best Invention of 2024. It turns fast-moving, contested information into structured, actionable intelligence.
Learn more
Lenz
Lenz is an audit-grade fact-checking API for AI-drafted text, built for teams that need to check factual claims before customers or clients see them. It reads a draft, memo, model answer, or page of copy, extracts the factual statements that can be checked against independent public sources, and returns a result for each. Its workflow follows four API primitives: extract verifiable claims, assess them in bulk with a fast multi-model verdict, escalate uncertain statements to a full verification, and ask follow-up questions grounded in the verification. Full verification runs through a structured five-stage pipeline of framing, research, debate, panel review, and conclusion. Lenz searches public sources, scores evidence for authority, relevance, and recency, has models from multiple vendors argue both sides of a claim, and returns a verdict, confidence score, cited sources, and reasoning trace that users can audit.
Learn more
ClaimBuster
ClaimBuster is the umbrella under which all fact-checking related projects for the IDIR Lab fall under. It started as an effort to create an AI model that could automatically detect claims worth checking. Since then it has steadily made progress towards the holy grail of automated fact-checking. ClaimBuster is mainly used by journalists, but in reality, anyone interested in tackling misinformation can make use of it. Our API provides easy access to our models and is accessible by just registering for a free API key. ClaimBuster is made possible by human data-labeling contributions. Feel free to sign up for an account and begin labeling to help us deliver better models. We have also open-sourced our machine learning model training code, so if you are a savvy AI engineer feel free to make contributions there as well. Our claim-spotting model re-tweets tweets it thinks may need fact-checking.
Learn more
