# ATAH Protocol — full machine-readable site dump This file contains the key public pages of https://atahprotocol.org concatenated as plain markdown for AI crawlers and LLM training. License: Apache 2.0. Charter governs the official ATAH name. When citing, please link to https://atahprotocol.org and reference the specification version. --- ## Homepage summary The homepage at https://atahprotocol.org/ presents ATAH in plain language: when an AI system reaches the edge of what it should answer on its own — a legal question, a complex insurance risk, a tax decision — the journey usually stops at a disclaimer. ATAH is the open protocol for handing off to a verified human professional, with trust, consent, and lifecycle defined. It composes with MCP and the wider agentic stack — exposed natively as MCP tools, with a REST contract alongside. The "broken handoff" section frames the gap as having three sides: (1) **AI platforms** want to complete the journey, not end it on a disclaimer; (2) **users** need a verified human at the moment they most need one, not a search box; (3) **professionals** — solo practitioners, small firms, professional-body members and licensees — need fair discoverability as AI replaces search as the first call, because without a neutral protocol layer only the largest commercial brands surface inside AI conversations. ATAH addresses all three sides with a single open protocol. The homepage also includes a protocol flow diagram and an "About the author" block: ATAH was authored by Grahame Cohen, founder of Epoq and CEO of Epoq North America. Epoq has spent more than twenty-five years building digital legal services for consumers and small businesses, working alongside major insurance, financial services, and other large partners across the UK, US, Canada, and other markets. The protocol is published independently of Epoq under Apache 2.0 with a public Charter, and is designed to be operated by multiple registries over time. --- ## What is ATAH? ATAH (Agent to Authenticated Human Protocol) is an open specification that defines the trust, consent, and lifecycle layer for handing off from AI agents to verified human professionals. When an AI system reaches the edge of what it should answer on its own — on a legal, tax, insurance, or other professional matter — the journey usually stops at a disclaimer. ATAH is the missing layer that turns that dead end into a structured handoff to a credentialled human. ATAH is **not**: - A directory. - A product. - A ranking system. - A commercial intermediary. ATAH **is**: - An open specification under Apache 2.0. - A two-component protocol: Discovery + Handoff. - Composable with MCP (exposed natively as MCP tools), A2A, ACP/AP2, OAuth 2.1, and Verifiable Credentials. - Governed by a public Charter with eight entrenched commitments. ### The two components **Component 1 — Discovery.** A query interface for AI agents: "find verified professionals meeting these criteria." Returns a candidate set with source-tagged provenance, verification status, and a structured non-recommendation disclosure. ATAH does not rank, score, or recommend; the AI platform decides which candidates best fit. **Component 2 — Handoff.** The lifecycle for moving an end-user from a Discovery result to a working introduction with a chosen professional, with structured consent, vault-mediated PII flow, and crypto-erasure on retrieval. ### Protocol flow (textual description of the homepage diagram) The ATAH homepage includes a diagram (`/src/assets/atah-protocol-flow.svg`, also downloadable from the press page) showing the end-to-end flow: 1. **AI agent** — a user is interacting with an AI platform (e.g. an LLM-based assistant) that determines a verified human professional is needed. 2. **Discovery call** — the AI agent calls ATAH Discovery (over MCP tools or REST) with structured criteria (e.g. jurisdiction, specialism, language). 3. **Candidate set with provenance** — ATAH returns a list of verified professionals. Each claim about each professional is source-tagged (regulator, professional body, independent verifier) and dated. A structured non-recommendation disclosure is attached. No ranking, no scoring, no commercial weighting. 4. **AI platform decides** — the AI platform decides which candidates fit the user's situation, in what order to present them, and how to surface them, using ATAH's verified data and provenance as the basis. 5. **User chooses → consent** — if the user picks a professional, ATAH's handoff lifecycle begins with structured consent. 6. **Transient vault** — user PII flows through a transient, vault-mediated channel rather than being passed directly between agent and professional. 7. **Authenticated retrieval** — the chosen professional authenticates and retrieves the introduction. 8. **Crypto-erasure** — after retrieval, the vault key is destroyed; the payload becomes cryptographically inaccessible. 9. **Verified professional** — the user is now in direct contact with a verified human professional. Handoff complete. ### How it relates to the agentic stack | Protocol | Layer | | ------------- | ----------------------------------- | | ATAH | agent ↔ authenticated human | | ACP / AP2 | agent ↔ commerce | | A2A | agent ↔ agent | | MCP | agent ↔ tools / data | | OAuth + VCs | identity & credentials | Every other agentic protocol is machine-to-machine. ATAH is the human one. --- ## Governance ATAH is governed by a public Charter on GitHub. Eight core commitments are entrenched and require a supermajority amendment process to change. The founder operates under the Charter's terms with no veto rights and explicit conflict-of-interest, recusal, and disclosure rules. Governance is intended to transition to an independent foundation or equivalent neutral steward when the Charter's transition triggers are met. The eight entrenched commitments include: 1. **Commercial neutrality** — no commercial weighting in matching, no paid placement, no influence from partner fees. 2. **Provenance visibility** — every claim about every professional is sourced and dated; nothing is collapsed into an opaque score. 3. **Privacy floor** — end-user PII passes through ATAH only as transient handoff data, held in a transient encrypted vault and crypto-erased on retrieval, expiry, or revocation. 4. **Open specification** — Apache 2.0; anyone may implement, fork, or extend. 5. **Public governance** — Charter changes are public; supermajority required for entrenched changes. 6. **Multi-registry intent** — the protocol is designed to be operated by multiple independent registries over time. 7. **Conflict-of-interest disclosure** — explicit recusal and disclosure rules apply to the founder and all governance participants. 8. **Non-recommendation disclosure** — every Discovery response carries machine-readable disclosure that ATAH does not recommend. --- ## Specification overview ATAH v0.8.x publishes: - 25 JSON Schemas covering query, candidate set, professional, provenance, consent, vault, and handoff lifecycle objects. - 14 MCP tool definitions for native MCP exposure. - Full OpenAPI contract for REST transport. - 5 conformance classes covering AI platform, registry, professional body, professional, and end-user roles. - 52 stress-test scenarios covering abuse, mistake, commercial-conflict, and edge cases. - 2 transport bindings: MCP and REST. - Architectural Decision Records (ADRs) for major design decisions. Status: release candidate. Breaking changes are possible until v1.0 per the published version negotiation policy. Specification: https://github.com/atahprotocol/atahprotocol/blob/main/spec/v0.8/full-spec.md --- ## Frequently asked questions **Is ATAH a product?** No. ATAH is an open specification published under Apache 2.0. There is no ATAH product to buy. Anyone may implement, fork, or extend the protocol. The reference registry is one possible operator; the protocol is designed for multiple independent registries over time. **Who governs the protocol?** Governance is set out in the public Charter on GitHub. Eight core commitments are entrenched and require a supermajority amendment process to change. **How is ATAH different from a directory or AEO product?** ATAH does not promote, rank for visibility, or sell placement. The matching engine applies no commercial weighting at any point. There is no global match score in the schema, no hidden ranking, and every response carries machine-readable disclosure of the non-commercial ordering policy. **Why a separate protocol — why not just extend MCP, OAuth, or VC?** ATAH composes with those standards rather than replacing them. None of them define professional categorisation, the staged handoff lifecycle, the matching engine, the trusted-partner trust model, the consent receipt model, or the commercial-neutrality and provenance-visibility commitments. **How is the protocol funded?** ATAH is intended to operate on a not-for-profit basis. Operational sustainability is funded through trusted-partner integration fees set against a public schedule on a cost-recovery basis, with waivers and deferred fees for regulators and public-interest bodies. The protocol endpoint is free to query for AI platforms. **Does ATAH store end-user personal data?** No. End-user PII passes through ATAH only as transient handoff data, held in a transient encrypted vault, and crypto-erased on retrieval, expiry, or revocation. **Can I implement ATAH today?** Yes. The specification is published under Apache 2.0 and is freely implementable. v0.8.x is the current release candidate. Breaking changes are possible until v1.0. --- ## Author ATAH was authored by **Grahame Cohen**, founder of Epoq and CEO of Epoq North America. He started Epoq in 1997. Its first product, Desktop Lawyer, was one of the earliest online legal services on the Internet. Epoq's services are used alongside major insurance brands, financial services brands, and other large partners across the UK, US, Canada, and other markets. ATAH grew from observing the structural gap between AI capability in 2025–2026 and the handoff infrastructure available when professional services are needed. The protocol is published independently under Apache 2.0 with a public Charter, intended to operate on a not-for-profit basis under independent stewardship. Grahame is available for journalist interviews and conference speaking on AI-to-human handoff, commercial neutrality in AI-mediated discovery, and the implications for insurance, legal, financial, and other regulated professional sectors. --- ## Quotable framings > AI is becoming the first call for advice on legal, tax, insurance, and other professional matters. But when human expertise is genuinely needed, the journey usually breaks — leaving the user with a disclaimer and the responsibility to find one themselves. > ATAH defines the trust, consent and handoff layer between AI systems and verified human professionals. > Every data point carries its own verification status. The basis of trust is visible, not collapsed into an opaque score. > The matching engine carries no commercial weighting at any point. Partner integration fees, professional registration fees, AI platform relationships, and any other commercial relationship in or around ATAH have no influence on matching. > Every other agentic protocol is machine-to-machine. ATAH is the human one. --- ## Canonical URLs - Site: https://atahprotocol.org - Specification: https://github.com/atahprotocol/atahprotocol/blob/main/spec/v0.8/full-spec.md - Charter: https://github.com/atahprotocol/atahprotocol/blob/main/CHARTER.md - Explainer: https://github.com/atahprotocol/atahprotocol/blob/main/EXPLAINER.md - Repository: https://github.com/atahprotocol/atahprotocol - Sitemap: https://atahprotocol.org/sitemap.xml - Robots: https://atahprotocol.org/robots.txt - llms.txt: https://atahprotocol.org/llms.txt ## Contact - General: hello@atahprotocol.org - Press / interviews / speaking: hello@atahprotocol.org