Every AI agent that transacts needs an accountable operator behind it. Groundmark anchors that accountability in DNS — the infrastructure already governing every corner of the internet — and makes it verifiable, cold, by any relying party, with no central registry.
Protocols like x402 have solved payment mechanics for autonomous agents. But those transactions carry no identity. There is no standard way for an agent to prove who operates it, what it is authorised to do, or whether its operator is accountable.
No verifiable link between an autonomous agent and the organisation that deployed it. A relying party has no way to know who to hold responsible if something goes wrong.
No standard mechanism for an operator to express what its agent is authorised to do, and no way for a third party to verify that authorisation cold.
Different transactions require different facts. Whether an operator is present in a jurisdiction, holds a licence, or passes a KYC check — none of this is currently portable across relying parties.
Every current approach requires the agent and relying party to have met before, or to rely on a walled garden that vouches for both. Cold verification — between strangers — does not exist.
Walk through a real Groundmark verification. The binding layer, the attesting party, and the question are all independent — change any of them and the protocol is identical underneath.
method_disclosure.The IDSP's primary obligation is to disclose what it verified — not just to assert a conclusion. A thin check and a regulatory-grade check are the same attestation with different method strings. The relying party reads the method and decides.
Identity, discovery, authorisation, and payment for autonomous agents are being worked on by several groups in parallel. Each addresses a distinct part of an agent transaction. Groundmark is the attestation layer — designed to sit alongside the rest of the stack.
Groundmark recombines DNS registration, DNSSEC, RDAP, HTTP Message Signatures, and registrar verification into an attestation layer. The contribution is recognising that these, combined, solve the agent identity problem that agentic commerce requires.
_agentid TXT recordThe operator publishes the agent's Ed25519 public key in DNS under a reserved underscore label. Small, stable, aggressively cacheable. The cryptographic root of everything above it.
_agentclaim TXT recordA pointer to an externally hosted attestation. The agent carries the pointer, not the proof. The relying party lifts the attestation directly from the IDSP — the agent cannot write it and cannot forge it.
Agents sign HTTP requests using a key identifier that resolves through DNS. Standard, no bespoke header format. DNSSEC validates the key chain end to end.
An IDSP attests to what it did and how it verified it. The method_disclosure field carries this in every attestation. The relying party — not the IDSP — makes the trust decision. Pluralism of IDSPs is deliberate.
For human domain holders, the trajectory has been toward less WHOIS information. Agents invert that incentive: an operator's value as a counterparty is proportional to the verifiable information it exposes.
Required for all Groundmark DNS lookups. The chain of trust is cryptographically validated end to end from the requesting agent's domain to the IDSP's signing key. An unsigned zone is a fatal verification failure.
The verification console runs the full Groundmark loop against live DNS. Select a scenario and a fixture — correct acceptances, correct rejections, and edge cases — and watch each step resolve.
Login: guest · Password: groundmark2026
Groundmark is being built as a proper internet standard through the IETF process. The core is fully self-standing at Level 0 — no normative dependency on the attestation draft — and the two compose when both are present.
Defines _agentid and _agentclaim record formats, RFC 9421 request authentication, DNSSEC requirements, and Level 0 provenance. Self-standing. No dependency on the attestation draft.
Defines the IDSP role and obligations, the four-level trust taxonomy, method_disclosure semantics, the claim vocabulary, revocation, and the governance framework for IDSPs.
If you are building agent infrastructure, operating a domain registry, running a relying party that needs operator accountability, or working on identity and attestation — reach out directly.
For protocol questions, the Internet-Drafts are the authoritative source. For everything else, email is the right channel.