- From: Christian Hommrich <christian.hommrich@trailprotocol.org>
- Date: Tue, 21 Apr 2026 11:53:19 +0200
- To: public-credentials@w3.org
- Cc: marcus@engvall.email, kyle@pryvit.tech, will@legreq.com
Hi all, Quick note before the thread moves on. I've been reading carefully these last days, and I want to thank Marcus, Kyle, and Will for the critique. It is helping me a lot to sharpen what our protocol needs to claim and what it should not. Transparency up front: I have a mechanical engineering background and currently build B2B AI sales systems — which is where I first ran into the agent accountability problem TRAIL tries to address. I started posting to this group in early April, so I am new here. English is not my first language, and I use LLMs to hit the right technical terminology and not skip the relevance of my arguments when writing in a group with this depth of expertise. A thread in my native language would help me, but it would not help the discussion. I could leave typos in, but that is something I care about avoiding, with or without AI. My reply this morning in the "LLMs and Agents usage in the CCG" thread shows several of the hallmarks Marcus described. That is a real pattern, not an accident. Kyle's point deserves a direct answer: identity credentials do not solve the mailing-list spam problem. He is right on the key-management angle, right on the puppeteer attacks, and the GitCoin sybil data is real. TRAIL is not built for that. We're working on a different problem - post-hoc accountability and revocation when AI agents are already acting in high-stakes operations: payments, audits, contracts, EU AI Act compliance. Not prevention of AI text in standards discussion. Different problem, different tool. For this mailing list, I align with Will - human-led norms beat technical filters. Identity layers do not belong here. Looking beyond this thread, though: AI-assisted communication is becoming the default across most professional contexts. That actually makes TRAIL's direction more relevant, not less - accountability and revocation tools become more important as AI agents take on more communication and more action. This is one of the reasons I started the protocol in the first place. I will keep working AI-assisted on the technical side, because the protocol needs that clarity to earn its vision. Being transparent about the method matters, and critique like this thread helps me draw the line between what TRAIL can legitimately claim and what it cannot. Happy to talk more about the difference between "preventing AI in standards work" and "making AI agents accountable when they act in the world" - those are different problems. Thanks, Christian
Received on Tuesday, 21 April 2026 09:53:38 UTC