Browsertrust forfrontierLLMs.
BIFROST adds deterministic verification signals directly on top of the AI tools people already use. ChatGPT, Claude, Gemini, Grok, Kimi, DeepSeek, MiniMax, Genspark: same discipline, same visible refusal boundary.

One browser layer across the tools people actually use.
The extension has one job: keep the answer honest.
Only model answers get evaluated. Prompts and user questions stay outside the verdict layer.
If the verifier is offline or the page cannot be parsed, BIFROST shows unavailable instead of green-lighting.
Hedges, missing sources, and ungrounded certainty lower the verdict. Confidence has to be earned.
When an answer violates the boundary, BIFROST surfaces the reason instead of smoothing it away.
Six deterministic engines. One visible verdict.
Normalize the model response
Extract claims and implied claims
Detect uncertainty and hedging
Probe for failure modes
Score confidence and risk
Return the visible verdict
The current install path is the Chrome extension build in the GitHub repo. The website points users straight to that extension section.
BIFROST stays public because the methodology is the point: visible checks, visible refusal, visible source.