Re: First post and related NeurIPS find

Thanks Erik for reply
and for the clarifications

I am reading and replying quickly, for fear that I may forget this
thread, until it becomes woven into something I am working on or
related

Basically I see a lot of relatedness distributed across efforts, so I
am attempting to find focus of what is essentially a kaleidoscopic
view of agentic AI

#
I am not sure I ll become more involved with the topic at this stage,
but  it is great to see people working alongside related dimensions
feel free to use the ideas I shared in the post and do not hesitate to
remind everyone what you are at

The IETF reference in my post is probably this?
https://datatracker.ietf.org/doc/html/draft-stone-aivs-00in
Does it sound right? *I forgotten on which list I exchanged with Ben
Stone about what he is doing

Sorry I sound vague

Look forward to be learning more

Best

PDM

On Wed, May 6, 2026 at 12:32 AM Erik Newton <eriknewton@gmail.com> wrote:
>
> Hi Paola,
>
>
> Thanks for taking the time to synthesize the stack. The three-layer framing you laid out (research, standards, implementation) maps well to how the pieces compose in practice, and I am glad you found the pattern.
>
>
> A few small clarifications from the Sanctuary side, since I am the framework author and can speak to that part directly:
>
> The AIVS archive format is still open. The CG kicked off recently and has not ratified an artifact format yet. The .aivs and .shm extensions you reference do not appear in any chartered AIVS work that I am aware of. Format selection is one of the things the CG will work through, and I would not want to anchor downstream readers on extensions that have not been agreed.
> The IETF draft reference puzzles me. I am not aware of an "Agentic Integrity Verification Standard" Internet-Draft. Do you have a datatracker link? It is possible I am missing something, or it may be a confusion with adjacent work. The W3C AIVS CG is not an IETF draft.
> DLA is a useful research anchor, not the AIVS charter. CRAS and Role Ontology are a good worked example of what role-adherence verification can look like, but AIVS scope is broader than DLA specifically. Other research will inform other parts of the spec.
> Sanctuary's Layer 3 Selective Disclosure ships redact and minimize patterns in v1.x. Zero-knowledge proofs of capability are a longer-horizon roadmap item, not current scope. Worth flagging because the ZK framing has been an attractor in the agent-identity space and people sometimes assume it ships sooner than it does.
> Sanctuary is one early implementation that can carry AIVS-conformant artifacts in its Layer 4 Verifiable Reputation surface. The CG is open to multiple implementations, and the broader composition pattern is bigger than any single framework.
>
> Beyond those clarifications, your composition framing is structurally sound and welcome. If you would like to engage further: the AIVS CG list is the right venue for the standards piece; for Sanctuary specifics the project lives at sanctuaryprotocol.ai with the source repo linked from there.
>
> Best, Erik Newton Sanctuary Framework author
>
>
>
>
> Erik Newton
>
>

Received on Tuesday, 5 May 2026 18:03:17 UTC