Observed arrival · 2026-10-07
Tellstone: a behavioral fingerprint for AI models
A browser-based checker compares an AI endpoint’s answers to low-stakes prompts against a trusted reference, looking for signs that the endpoint may be serving a different model.
- For
- Teams checking AI model routing and API endpoints
- Worth noticing
- Its one-endpoint Unmask mode compares against stored model records, though the page says a live reference is stronger.
Field notes
Each check sends eight low-stakes prompts—such as choosing a colour, naming a city, or flipping a coin—to a doubtful endpoint and a trusted reference, then compares their characteristic answers. The page also describes a one-endpoint Unmask mode that checks against stored model records, while noting a live reference is stronger. Its trial section says it staged cheaper-model, older-version, and cross-vendor substitutions alongside honest pairs, and identifies cases the detector cannot reliably catch.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue