- From: Tymofii Pidlisnyi <tima@aeoess.com>
- Date: Wed, 4 Nov 2026 12:21:46 -0600
- To: llewis@sovrlabs.com
- Cc: chgaowei@gmail.com, abisheakp197@gmail.com, keotheodangcha548@gmail.com, public-agentprotocol@w3.org
- Message-ID: <CAH0g2+J+2wOUCOTxUNh2ax1d5b-VdP4LL744ZGgXz4PC9QspjQ@mail.gmail.com>
Hi Larry, I mapped the four cases against the lab. It already tests actions outside a grant's scope and revocation of delegated authority. My draft candidate #158 tests whether a principal binding is associated with the right authority root, but assumes both records already verify: https://www.google.com/url?q=https://github.com/Agent-Authority-Conformance/aps-conformance-suite/pull/158&source=gmail&ust=1793902905423000&sa=E The lab doesn't yet test a credential that links a legal principal to a DID, whether its issuer is trusted for that claim, or whether the binding itself has been revoked. Revoking the delegation is a different test. I'd use a valid binding and an in-scope action as the control, then change one condition in each of the other three cases, an untrusted issuer, a revoked binding and an out-of-scope action. Do you have a credential profile and revocation mechanism in mind for the first set? Best, Tymofii On Mon, Nov 02, 2026 08:07 PM, Larry Lewis <llewis@sovrlabs.com> wrote: > Hi Gaowei, Ren, Abisheak, and all, > > > Just want to add my input. > > Gaowei, this is an important question, and I think it highlights a > distinction we should make explicit in the architecture. > > A DID establishes an identifier and supports cryptographic verification of > key control. It does not, by itself, establish the legal identity of the > person or organization responsible for that DID. > > I believe we should separate three concerns: > > 1. > > *Identity assurance:* An accepted issuer verifies the legal person or > organization and issues a signed credential establishing the binding to the > relevant DID. > 2. > > *Delegated authority:* The principal issues a verifiable authorization > defining what the agent is permitted to do, including scope, duration, > constraints, and revocation. > 3. > > *Runtime enforcement:* The receiving system verifies the identity > evidence, delegation chain, and current authorization before allowing a > consequential action. > > This does not necessarily require a single globally authoritative > institution. Different environments may recognize different issuers, > including governments, enterprises, and identity providers. The verifier > must still determine whether the issuer is trusted for the particular claim > being made. > > I think this aligns naturally with the interoperability work across ANP, > NTI, and Agent Trust Protocol (ATP). > > Rather than creating competing identity systems, we could define a > method-neutral principal-binding verification profile using existing W3C > DID and Verifiable Credential mechanisms. > > I would propose a small conformance test covering a valid principal > binding, an untrusted issuer, a revoked binding, and an agent attempting an > action outside its delegated authority. > > This would allow us to move from architectural discussion to independently > reproducible verification evidence. > > Best regards, > > Larry Lewis > Agent Trust Protocol Community Group > > > > From: Gaowei Chang <chgaowei@gmail.com> > To: "Abisheak P"<abisheakp197@gmail.com> > Cc: "Ren Yushan"<keotheodangcha548@gmail.com>, "public-agentprotocol"< > public-agentprotocol@w3.org> > Date: Mon, 02 Nov 2026 21:49:52 -0600 > Subject: Re: Subject: Re: [public-agentprotocol] Layer 0 Trust & > Verification > > That’s a very good idea. I have one question: how can a DID be linked to > an individual’s legal identity? How can developers trust this binding > relationship? Does this require a pre-established authoritative institution? > > On Sun, Nov 1, 2026 at 6:37 PM Abisheak P <abisheakp197@gmail.com> wrote: > > Hi Gaowei, Ren, all, > > Gaowei, you have hit the exact architectural crux. You are absolutely > right: Layer 0 (the underlying transport and identity root) is a > prerequisite. If we cannot establish trusted domain names and DIDs, secure > communication collapses. > > Building Layer 0 on top of HTTPS is the correct industry direction. > However, as you noted, establishing a DID only proves the identity of the > agent. It does not prove that the agent is authorized to perform a specific > action. > > This is the gap that NTI (Neutral Trust Infrastructure) addresses. We do > not compete with Layer 0 or DIDs. We anchor to them. > > Here is how NTI handles the second type of trust problem: > > 1. Principal-Bound Identity: Following Alan Karp's feedback, we bind the > DID of the agent to the responsible legal party (the human or enterprise). > 2. Runtime Non-Repudiation: Even with a trusted DID, an agent can be > hijacked by prompt injection. NTI acts as a runtime sidecar. It verifies a > post-quantum (Dilithium5) signature on every tool call. If the signature is > missing or invalid, it fails closed. > 3. The incomplete_check: If the revocation registry (where we check if a > DID has been compromised) is unreachable, we do not proceed. The agent > enters a PAUSE state. > > In short: Layer 0 proves the agent exists. NTI proves the agent is allowed > to execute this exact action at this exact moment. > > I would love to see how we can integrate the ANP identity layer with the > NTI enforcement layer. I am dropping the spec link here for your review > (and for anyone else asking the Layer 0 question): > > 👉 https://github.com/abisheakp197/nti-spec > > Best, > Abisheak > > On Sun, 1 Nov, 2026, 2:51 pm Gaowei Chang, <chgaowei@gmail.com> wrote: > > Let me briefly share my thoughts for discussion.我简要分享一下我的想法供大家讨论。 > > First, on trust. What is encoded in protocols may not be the trust we > speak of in everyday terms, but rather cryptographic security mechanisms. > Take HTTPS as an example: if I know the counterparty’s domain name, a > cryptographic mechanism can establish a secure, hard-to-compromise > connection between us. This is the foundational network trust enabled by > cryptography. It comes with a prerequisite, though: you need a trusted > domain name.首先谈谈信任。协议中所编码的或许并非我们日常所说的信任,而是密码安全机制。以 HTTPS 为例:若我知晓对方的域名, > 一种密码安全机制便能在我们之间建立安全且难以被破解的连接。这是密码学所支撑的基础网络信任。不过,这有一个前提条件:你需要一个可信的域名。 > > Another approach is something like Verifiable Credentials (W3C VCs): a > trusted authority issues a credential for a DID. By verifying that > credential, I know the DID has at least been attested by that trusted > authority. The trust placed in the authority itself may, in turn, stem from > offline arrangements or built-in machine trust roots.另一种方法类似于可验证凭证(W3C > 可验证凭证):受信任的机构为去中心化标识符(DID)签发一份凭证。通过验证该凭证,我就能知道该去中心化标识符( > DID)至少已得到该受信任机构的认证。而对该机构本身的信任,又可能源自线下约定或内置的机器信任根。 > > My point is that protocols may not fully solve this second type of trust > problem. If anyone has a strong solution, I’d welcome you to share it. > 我的观点是,协议或许无法完全解决第二类信任问题。如果有人有更好的解决方案,欢迎分享。 > > As for whether trust can be established using text, I’m not yet clear on > your exact line of thinking. Agents are indeed highly capable of processing > text. That said, for two agents to communicate, there must be a Layer 0 > protocol beneath the application layer; otherwise, secure communication > cannot be established at all.至于能否通过文本建立信任,我目前还不清楚你具体的思路。 > 智能体确实具备极强的文本处理能力。但即便如此,两个智能体要进行通信,应用层之下必须存在0层协议,否则根本无法建立安全通信。 > > The only real question is where this Layer 0 should sit. The prevailing > view across the industry today is to build the Layer 0 protocol on top of > HTTPS, which is also the direction we are taking.唯一真正的问题是这个第0层应该位于何处。 > 目前整个行业的主流观点是在HTTPS之上构建第0层协议,这也是我们正在采取的方向。 > > On Thu, Oct 22, 2026 at 7:03 AM Ren Yushan <keotheodangcha548@gmail.com> > wrote: > > Hi all, > > I'd like to raise a point about one interaction in the ANP protocol — > specifically, how trust between agents is handled once they are talking to > each other. > > If two agents want to communicate, they first have to discover each other > and establish a connection. A separate concern is that, while they are > chatting, they need to trust each other. > > The difficulty arises when that trust is written into the protocol. It has > to become a normative statement — a formulation you use to verify that the > other party is a safe object. But in reality, when we chat, there is no > security-verification step. We do not run some protocol while talking in > order to guarantee that the information is safe. Safety is something we > discover during the conversation: whether this person is reliable or not. > > So I would argue that the protocol should not try to define what is safe, > what is unsafe, and what is trustworthy. Instead, it could offer reference > suggestions about what means agents might use to communicate — for example, > a password, or a standardised JSON format. But if that format is hard-coded > into the protocol, it cannot be switched; and if it later turns out to be > insecure, the protocol itself acquires a hole. > > So could we take a step back: not define the protocol, only give reference > definitions? Since agents can communicate in text, why not let them use > text to establish trust? Give them a verification method they can choose — > they decide whether to use it, and which one. That is my view of trust. > Happy to hear counterarguments, especially from anyone who thinks trust can > be made normative without the protocol becoming brittle. > > Best regards, > Yushan Ren > > > >
Received on Wednesday, 4 November 2026 18:21:52 UTC