Re: Introduction + AI Agent Auditability Proposal

Dear Ben and everyone
Thanks for sharing

While I am teaching myself to use github, respec and a bunch of other
things going around my head and the web
and trying to fix them in some form https://w3c-cg.github.io/aikr/

I have consulted with my oracles and gathered some thoughts on your spec
https://w3c-cg.github.io/aikr/conduit/index.html

Please check, let me have feedback as to what makes sense or not and
edits/comments via PR while I get my head around this and other things

Best

Paola


On Tue, Mar 17, 2026 at 8:59 PM Ben Stone <benstone@swarmsync.ai> wrote:

>  Hi everyone
>
> I am Ben, a developer who is working on AI agent infrastructure. I
> recently joined this community group. I wanted to introduce myself.
>
> I have been building a tool called Conduit. Conduit is a browser that
> creates a tamper-proof audit trail of everything an AI agent does on the
> web. The core idea of Conduit is that after an AI agent session you can
> hand someone a file, and they can verify exactly what the AI agent did
> without trusting any server or third party.
>
> As part of that work I wrote a specification called the Conduit Session
> Proof Format. The Conduit Session Proof Format is a proposed standard for
> how AI agent sessions should be documented and verified. The Conduit
> Session Proof Format is designed to satisfy things like the EU AI Acts
> audit log requirements with a interoperable format.
>
> I think there is a question in the AI agent space around accountability.
> How do we prove what an AI agent did? I would love to contribute to that
> conversation about AI agent accountability
>
> The Conduit specification is available, on GitHub:
> https://github.com/bkauto3/Conduit
>
> I am happy to be here
> Ben
>
>