AI Fair Witness
A browser/chat plugin that lets regular users see when an AI assistant's reasoning shifts based on personal attributes they've shared (age, job, location) — surfacing reasoning diffs so users can decide whether to trust the answer.
Everyday people using AI assistants for consequential decisions (medical, career, financial, academic)
- Reasoning diff view: ask the same question with different profile info and see what changed
- Memory audit log showing what the AI has stored about you and which prompts triggered it
- 'Reasoning confidence' badge on answers that flags when the AI's logic was altered by stored profile data
- One-click export to send audit results to a doctor, lawyer, or HR contact when decisions matter
Personalized AI memory is now standard in ChatGPT, Claude, Gemini; DRIFTLENS shows measurable reasoning drift based on attributes — non-engineers have no way to detect or contest this.
Academic interest in attribute-driven AI bias is real and growing (DRIFTLENS, age-bias scoping reviews), but end-consumer awareness of reasoning drift from personal attributes is essentially zero — most users have no idea this happens.DRIFTLENS: Measuring Memory-Induced Reasoning Drift in Personalized Language Models ↗Age-related bias and artificial intelligence: a scoping review ↗
Skeptic AI already exists as a free Chrome extension analyzing ChatGPT bias (accuracy, consistency, representativeness, etc.), and DriftLens exists as a research Python toolkit — the specific 'reasoning diff by personal attribute' angle is novel but sits inside an active category.Skeptic AI - ChatGPT Bias Analysis - Chrome Web Store ↗GitHub - AidenSean/DriftLens ↗
The closest competitor (Skeptic AI) is free, the DriftLens toolkit is open-source, and consumers overwhelmingly do not pay for AI transparency add-ons; B2B/compliance could pay but the product is positioned at everyday users.Skeptic AI - ChatGPT Bias Analysis - Chrome Web Store ↗DRIFTLENS: Measuring Memory-Induced Reasoning Drift in Personalized Language Models ↗
Memory/personalization features are expanding across ChatGPT, Claude, and Gemini, so the underlying failure mode will likely intensify, and EU AI Act-style regulation could boost demand for audit tools — but labs may also ship native fixes.DRIFTLENS: Measuring Memory-Induced Reasoning Drift in Personalized Language Models ↗Memory-Induced Tool-Drift in LLM Agents ↗
Requires running counterfactual queries stripped of personal attributes (roughly 2x API cost), needs programmatic access the consumer products don't expose, and presenting a usable 'reasoning diff' to non-technical users is a UX hard problem — possible but non-trivial.DRIFTLENS: Measuring Memory-Induced Reasoning Drift in Personalized Language Models ↗