Capture AI conversations as evidence

AI assistants are now where real work happens: contract language gets drafted, medical questions get asked, code gets generated, decisions get shaped. When any of that is later disputed — what exactly did the assistant say? what was it asked? when? — a screenshot proves nothing. The Sigill browser extension turns AI conversations into evidence at the moment they happen.

What it does

Once installed, the extension works in the background on Claude, ChatGPT, and Gemini:

• Every conversation turn — prompt and response — is captured into a canonical evidence envelope (RFC 8785 JSON) and timestamped via RFC 3161 as it occurs, not reconstructed afterwards.
• Content is hashed on your side; the timestamping is hash-based, like everything on the platform.
• Sealed conversations appear in your Evidence view, tied to your organisation's tenant, where they can be exported and independently verified later.

Because each turn is timestamped as it happens, the evidence has a property no export-after-the-fact can match: it is tamper-evident from the moment of the conversation, and each turn's proof is independently verifiable with standard RFC 3161 tooling — see Verify a seal without a Sigill account.

Who reaches for this

• Compliance and legal teams establishing what AI systems actually told employees — or what employees actually asked.
• Professionals relying on AI output (advice, analysis, drafting) who may later need to show precisely what the assistant produced, unaltered.
• Anyone establishing priority: a conversation where an idea first appeared, provable to the minute — the conversational sibling of proving priority for IP.

Setup

Add the extension from the Chrome Web Store — the only official distribution channel — then visit sigill.ai while signed in and the extension picks up your identity automatically. Permissions, data handling, and the privacy policy are on the browser extension page; the launch post shows a full capture-to-verification walkthrough.

The rest of the family

The extension covers the human-using-an-assistant case. For programmatic AI evidence, the same envelope model is available in code: AI evidence envelopes via the SDKs for individual model calls, JAdES seals for agent logs, and MCP for agents that seal their own work. One evidence model, four capture points.