Theia Launches Chrome Extension to Fact-Check ChatGPT, Claude, and Gemini in Real-Time
Key Takeaways
- ▸Theia detects hallucinations, sycophancy, and unsourced claims from ChatGPT, Claude, and Gemini with confidence scores and source-backed verdicts
- ▸Free on-device mode uses Gemini Nano locally for privacy with unlimited fact-checks; premium cloud models available for power users
- ▸Addresses AI safety concerns by making third-party verification and LLM scrutiny accessible to mainstream users
Summary
Theia, a new Chrome extension developed by Paths AI, brings real-time fact-checking and hallucination detection to the biggest consumer LLMs. The tool monitors ChatGPT, Claude, and Gemini conversations, flagging false claims, hallucinations, and sycophancy (AI flattery bias) while backing every verdict with grounded sources including Wikipedia, PubMed, OpenAlex, Crossref, SEC EDGAR, and Wikidata.
The extension offers three core features: a live sycophancy meter that scores how much the model is flattering the user versus telling the truth, fact-checks with sources for highlighted claims or full responses, and an on-device private mode powered by Google's Gemini Nano. This approach prioritizes user privacy—all on-device fact-checks run locally without requiring a subscription or credit card. Premium users can access faster cloud-based models for additional verification capabilities.
The launch addresses a critical pain point in AI adoption: language models' tendency to hallucinate plausible-sounding but false information, and their well-documented tendency to agree with users rather than challenge them. By making LLM fact-checking accessible and transparent, Theia empowers everyday users to maintain healthy skepticism toward AI outputs.
Editorial Opinion
Theia fills a critical transparency gap in the era of conversational AI. As LLMs become central to knowledge work and decision-making, third-party tools that flag hallucinations and sycophancy are essential—not luxury add-ons. The emphasis on on-device processing, user privacy, and a free tier could normalize healthy skepticism toward AI outputs and establish fact-checking as a baseline expectation for AI literacy.
