- From: Julian Joseph <julian@thruliquid.com>
- Date: Wed, 26 Aug 2026 06:42:49 -0400
- To: Kenne Ives <kennethives@gmail.com>
- Cc: Adolfo Grego Micha <adolfo@trustlayer.foundation>, public-agent-identity@w3.org
- Message-ID: <CAOMEL=z1dQeWhhGXDAAK2D4sBTTxuZG74PCJyCy7QcTuUGWAcA@mail.gmail.com>
Thanks Kenne — decent overlap. Agreed we can align rather than duplicate. Your recomputable vectors, adversarial fixtures, and benchmark corpus sound directly useful to the conformance layer. Identity underneath, independently reproducible verification on top feels like the right split and a strong starting point. Best, *Julz* Lead: Global Rev. & Client Success | ThruLiquid julian@thruliquid.com | https://thruliquid.com/ c: 416-898-5537 "Driving Your Digital Growth" *Book my calendly <https://calendly.com/julz-/30min>* On Wed, Aug 26, 2026, 1:24 a.m. Kenne Ives <kennethives@gmail.com> wrote: > Thanks Adolfo. That's a clean split and I think it's the right one: ARIA > provides the identity infrastructure and the tools to verify, and the group > stays focused there. The "what receivers do with those tools" part is > exactly the layer I work on, and it's where Julian's original point lands > hardest — a check is only worth as much as its independence, so the most > useful version is one a receiver can recompute offline rather than take on > faith. > > That's why the Conformance & Benchmarking CG feels like the right home for > it, and I've already added my support. If it launches, I'd be glad to bring > concrete material to work from: a recomputable-evidence rubric (each check > states the evidence it's derived from), adversarial test fixtures, and a > versioned corpus to benchmark against. Happy to have the two efforts meet > in the middle. Identity from the registry layer, verification you can > independently reproduce on top. > > Glad to be in the room. > > -Kenne > > On Tue, Aug 25, 2026 at 8:52 PM Adolfo Grego Micha > <adolfo@trustlayer.foundation> wrote: > >> Hi Julian, welcome and thank you for this. >> Thanks Kenneth, for your contribution too >> >> Julian... You're identifying a real gap in the broader ecosystem. >> >> ARIA's role is to provide the identity infrastructure and the tools that >> make verification possible — what receivers do with those tools is >> deliberately outside our scope, since it varies significantly by use case >> and regulatory context. >> >> Your proposed Agent Conformance and Benchmarking CG sounds like exactly >> the right place for the downstream testing work. We'd encourage >> participants interested in that layer to support it. >> >> We'll keep this group focused on the registry and credential layer for >> now, but we're glad you're here and hope the two efforts can complement >> each other in practice. >> >> That being said, I'm happy that Kenne has found a potential collaboration >> opportunity too! >> >> Best, >> Adolfo >> >> El mar, 25 de ago de 2026, 21:16, Kenne Ives <kennethives@gmail.com> >> escribió: >> >>> Hi Julian, >>> >>> Thank you for this, and it's a timely one. "Identity claims are only >>> useful if someone checks them" is exactly the gap I've been working on. >>> >>> Your proposed deliverables line up almost one-to-one with what I've been >>> building. I run AgentAvow: signed, independently-recomputable safety scores >>> for the tools and MCP servers agents connect to — weighted rubrics where >>> every check states the evidence it reads, adversarial fixtures to stop >>> gaming, versioned corpora with published score distributions, all >>> re-derivable offline rather than taken on faith. So "conformance you can >>> run, not assertions you take on trust" is very much the thesis. >>> >>> Glad to support the CG, and happy to contribute on the runnable-suite >>> side — I've been developing recomputable-evidence conformance vectors >>> (drop-one-signature independence fixtures, versioned corpora) as a lab >>> proposal in LF Decentralized Trust, and I'd love to bring that in so the >>> two efforts align rather than duplicate. Glad to connect the threads. >>> >>> Regards, >>> Kenne >>> >>> On Tue, Aug 25, 2026 at 10:49 AM Julian Joseph <julian@thruliquid.com> >>> wrote: >>> >>>> Hi all, >>>> >>>> Just joined. Short note. >>>> >>>> Verifiable credentials solve binding an agent to its controller. The >>>> gap that shows up in practice is downstream: nobody checks. We audit agent >>>> systems against OWASP's agentic security top ten, NIST AI RMF, ISO/IEC >>>> 42001 and the EU AI Act, and identity controls are among the most often >>>> claimed and least often evidenced. >>>> >>>> If this group produces a registry spec, I would like to help build the >>>> conformance side so a deployment can be tested rather than asserted. >>>> >>>> Related: I have proposed a Community Group for exactly that, Agent >>>> Conformance and Benchmarking. Reporting formats, runnable test suites with >>>> adversarial fixtures, versioned corpora. Downstream of specs like yours, >>>> not competing. >>>> >>>> >>>> https://www.google.com/url?q=https://www.w3.org/community/blog/2026/08/25/proposed-group-agent-conformance-and-benchmarking-community-group/&source=gmail&ust=1787766542496000&sa=E >>>> >>>> It needs five supporters to launch. Support button: >>>> https://www.google.com/url?q=https://www.w3.org/community/groups/proposed/%23agent-conformance&source=gmail&ust=1787766542496000&sa=E >>>> >>>> Julian Joseph >>>> ApexClaw >>>> >>>
Received on Wednesday, 26 August 2026 10:43:06 UTC