For SDOs

This page is written for engineering Standards Developing Organizations (SDOs) and their advisors: general counsel, policy leads, standards-development volunteers, and publishing staff who set the terms under which the organization’s content reaches the field. It is not a sales page. It is an attempt to explain, in roughly the time it takes to read a long memo, what CLASP is for, what an SDO would do to adopt it, and what risks it does and does not address.

The problem prohibition does not solve

The dominant SDO response to agentic AI has been to prohibit it. ASTM’s terms state that using artificial intelligence on ASTM standards and related intellectual property is prohibited, with violations resulting in suspension of access. ISA has taken a similar position. IEEE has rewritten its developer API terms to define AI, ML, and LLM as distinct categories with significant restrictions. ASME, IEC, ASHRAE, and NFPA have not published explicit AI policies, but their subscription terms forbid redistribution and derivative use in ways that most reasonable interpretations extend to AI ingestion. The shared posture is “no.”

The intent behind the posture is legitimate. SDOs have two reasonable fears: that AI ingestion amounts to uncompensated bulk redistribution of standards content, and that AI-mediated summaries will displace the subscription revenue that funds the standards-development work itself. Both fears have a real kernel.

The posture does not, however, achieve its purpose. Practicing engineers increasingly work alongside agentic tools whether their professional society endorses it or not. An engineer who cannot consult a standard inside an AI-assisted design session does one of three things: works around the prohibition (anonymous web extraction, screenshotting, asking the model to recall what it already absorbed), proceeds without consulting the standard at that moment, or assumes the model already knows. None of those outcomes serve the SDO, the engineer, or the field. The SDO loses visibility into use; the engineer ends up working from whatever the model happened to absorb at training time, often years stale; and the standard’s authority is quietly displaced by an unattributed summary. Prohibition does not prevent AI access to standards content. It prevents licensed, traceable, citable AI access.

The architectural answer is additive

CLASP, the Clause-Level Access and Standards Protocol, is an open specification (CC-BY-4.0) that extends the IAB Tech Lab’s Content Monetization Protocols (CoMP) v1.0 for the domain of engineering and technical standards. The IAB Tech Lab is the non-profit consortium that finalized CoMP on April 28, 2026; it has a track record of widely-adopted open specifications in adjacent industries (OpenRTB, ads.txt, VAST). CoMP defines the handshake by which an AI System and a Content Owner agree on commercial terms before content is accessed. CLASP populates CoMP’s extension placeholders with engineering-specific vocabulary and adds a free discoverability tier that sits adjacent to the handshake.

CLASP is strictly additive to existing SDO infrastructure. The SDO’s existing subscription platform continues to serve human users with the access modes it serves today (PDF, HTML, search, annotation). New CLASP endpoints serve licensed AI agents, alongside, on the same domain, against the same back-office license registry. The revenue model is unchanged. The access mode is expanded. No part of CLASP requires an SDO to discount licenses, publish content for free, or replace existing distribution channels.

What an SDO actually deploys

Four components, each on its own narrowly-scoped endpoint. Each is small enough that an SDO’s existing platform team can stand it up; nothing here implies replatforming.

CoMP front door. A single HTTPS endpoint that accepts a CoMP handshake request from an AI System (identifying itself, declaring its intended use, naming the license it claims) and returns a CoMP Package response (offered scope, retrieval endpoint, authentication requirement, citation requirement). The SDO’s existing license registry is the source of truth for who is allowed in; the front door is a thin protocol adapter, not a new licensing system.
MCP retrieval endpoint. A Model Context Protocol server exposing six clause-level tools: get_clause, search, resolve_term, list_changes, resolve_reference, compliance_check. The endpoint is authenticated (OAuth2) and gated by the same license registry as the front door. Tool semantics are normative: get_clause returns the exact clause text of the specific edition requested; search returns snippets, not full text; list_changes returns structured diffs between editions; compliance_check returns structured findings without paraphrasing normative text. AI agents cannot bypass the tool semantics by asking nicely.
Discoverability catalog. A free, unauthenticated, HTTP-cacheable catalog at https://<publisher-domain>/.well-known/clasp-catalog. It exposes titles, scope statements, edition history, and tables of contents. It does not expose normative clause text, defined-term text, or anything else the SDO considers normative. The catalog’s job is to let an AI agent recognize that a relevant standard exists and direct an unlicensed user toward the licensing flow; it is the SDO’s storefront, not a free read.
Citation envelope on every response. Every retrieval response from the MCP endpoint carries a structured citation envelope: publisher, designation, edition, language, retrieval timestamp, license fingerprint, and per-clause content hashes. The envelope is the artifact that downstream deliverables attach to the cited material; it is what an AI agent presents to its user, to a reviewer, or to a regulator to prove that a normative claim is backed by a real retrieval against the SDO’s real endpoint.

Two of these components (the front door, the MCP endpoint) plug into the SDO’s existing license registry. One (the catalog) is a thin static or near-static publishing surface. One (the citation envelope) is a wire format that the MCP endpoint produces. None of them touches the editorial pipeline, the standards-development workflow, or the human-facing subscription product.

The three risks that matter to an SDO

The risks worth talking about are the ones that actually keep an SDO’s general counsel up at night. CLASP addresses each directly.

Loss of attribution

The standard fear is that AI mediation strips authorship. An engineer asks an assistant what the wash-water requirements are; the assistant gives an answer; the engineer’s deliverable does not credit the SDO. Over time, the standard becomes invisible inside its own field of practice.

CLASP’s answer is the citation envelope. Every retrieval response carries publisher, designation, edition, clause identifier, language, and a content hash. AI System obligations are normative: an AI agent emitting CLASP-retrieved content into a downstream deliverable MUST attach the envelope. An AI agent that omits the envelope is non-conformant, and the SDO can refuse to renew the license. The standard’s authorship is enforced at the protocol level, not left to the AI vendor’s good intentions.

Bypass of subscription

The second fear is that an AI agent will use the protocol to extract content at a rate or in a form that effectively substitutes for a subscription, displacing the revenue model.

CLASP’s answer has three parts. First, retrieval is OAuth2-gated; an AI agent without a current license gets nothing past the front door. Second, the SDO retains revocation; if an AI agent abuses the channel, the SDO ends the session immediately, and downstream deliverables produced before revocation continue to carry their envelopes for traceability. Third, the tool semantics are designed to prevent bulk extraction in disguise: search returns snippets with bounded offsets, compliance_check returns structured findings without paraphrasing normative text, and get_clause is per-clause by design. Pricing, metering, and rate-limiting remain the SDO’s commercial decisions; CLASP does not constrain them.

Hallucination of fake clauses

The third fear is the most underappreciated and, in practice, the most damaging. An AI model invents a clause that does not exist, attributes it to an SDO’s standard, and a downstream deliverable carries the fabrication into production. The SDO’s reputation absorbs the cost of an error it did not make.

CLASP’s answer is mode = 0 exact-text citations. An AI agent emitting a verbatim quotation MUST have a mode = 0 citation present in its envelope, and the content hash MUST match the retrieved text. An AI agent claiming a clause-by-clause gap analysis MUST have mode = 4 structured-analysis citations. An AI agent claiming a clause changed between editions MUST have mode = 3 diff citations. There is no mode for “I think this is what the clause says.” If the citation is not present, the AI agent is contractually forbidden from emitting the content, and an SDO’s counsel has a concrete instrument to point to in an enforcement conversation with an AI vendor.

What CLASP is not

CLASP is not a clearinghouse. It does not issue licenses, hold money, or arbitrate disputes between SDOs and AI vendors. The SDO sets terms and the SDO collects revenue, exactly as today.

CLASP is not a marketplace. It does not aggregate standards across publishers, does not surface comparisons between competing standards, and does not direct traffic between SDOs. Each SDO’s catalog lives on the SDO’s own domain.

CLASP is not a way to discount licenses. The protocol is silent on price. An SDO can charge whatever the market will bear; the protocol carries no expectation that AI access is cheaper, more expensive, or differently structured than human subscription access.

CLASP is not an obligation to publish for free. The discoverability catalog exposes metadata, not normative text. An SDO can decline to expose any catalog at all and operate CLASP only on a direct-relationship basis with named AI vendors.

CLASP is not vendor-specific. It is an open specification, licensed CC-BY-4.0, built on an IAB Tech Lab specification licensed under Creative Commons. There is no Anthropic, OpenAI, or Google dependency. Any AI System that implements the CoMP handshake and MCP can interoperate.

FAQ

Does this require us to make our content free?

No. CLASP gates retrieval behind OAuth2 and the SDO’s existing license registry. The free tier is metadata only (titles, scope statements, tables of contents); normative text never appears in the unauthenticated catalog.

What happens to our subscription revenue?

The subscription model is preserved. Human users continue to use the existing subscription platform. AI agents access content through a separate licensed channel that the SDO prices independently. Most early SDO adopters will likely set AI-agent licensing as an add-on to existing institutional subscriptions, but the protocol does not constrain that choice.

What if we want to revoke an AI agent’s access?

Revocation is an SDO action and takes effect immediately. The protocol specifies that the AI System’s current session ends; new tool calls fail. Citation envelopes already attached to downstream deliverables remain valid as historical records, which is the right outcome for audit and traceability. An engineer’s previously-produced design package does not silently lose its citations when an AI vendor falls out of good standing with the SDO.

How is this different from existing API access we already provide?

Existing SDO APIs (IEEE Xplore’s developer API, ASTM Compass programmatic access where it exists) are designed for institutional integrations such as library systems, citation managers, and document-delivery platforms. They presuppose a human reader on the receiving end. CLASP is designed for the case where the receiving end is an AI agent acting on a human’s behalf, with the additional discipline that the AI agent’s downstream output is structured to carry citation back to the SDO. The handshake declares intended use; the tool semantics constrain what the AI agent can produce; the envelope enforces attribution.

Why should we trust IAB Tech Lab’s protocol?

The IAB Tech Lab is a non-profit standards-setting consortium with a 25-year track record of producing widely-adopted open specifications (OpenRTB, ads.txt, VAST, Open Measurement SDK). CoMP v1.0 was finalized April 28, 2026 as the output of a multi-stakeholder working group and is licensed under Creative Commons. CLASP is the engineering-domain extension, not a replacement; if the IAB Tech Lab’s pattern fails, the engineering domain’s specification can be re-anchored without breaking SDO deployments, because CLASP versioning is independent.

What about standards incorporated by reference into regulation?

This is the case that requires the most care. When a standard is incorporated by reference into a regulation, a specific edition is frozen for regulatory purposes even after the SDO publishes newer editions. CLASP’s scope.ext carries explicit edition identifiers and a RegulatoryIncorporation object that names the incorporating jurisdiction and rule. An AI agent producing a regulatory-submission deliverable retrieves against the edition the regulator cites, and the citation envelope records which edition was actually retrieved. The SDO’s choice of which editions remain retrievable through CLASP is an SDO choice; the protocol simply records it accurately.

Is this an Anthropic, OpenAI, or vendor-specific initiative?

No. CLASP is an open specification licensed CC-BY-4.0, layered on an open IAB Tech Lab specification licensed CC. The protocol uses MCP, which is itself an open standard with implementations across vendors. There is no commercial dependency on any particular AI vendor.

What does a pilot deployment look like?

A typical pilot scopes to a single edition of a single standard, exposed through a CoMP front door and an MCP endpoint, with the discoverability catalog showing the standard’s metadata. A small number of named AI vendor partners receive pilot licenses against the SDO’s existing license-issuance process. Telemetry from the MCP endpoint feeds back to the SDO’s usage analytics. The pilot runs against a real edition with real engineers, not against synthetic content. A reasonable pilot is six to twelve weeks from kickoff to first real retrieval; most of that time is the SDO’s existing internal review processes, not protocol implementation.

A note on the acronym

CLASP here refers to the Clause-Level Access and Standards Protocol. It is unrelated to CLASP, the Collaborative Labeling and Appliance Standards Program (clasp.ngo), which has worked on appliance energy-efficiency standards since 1999. The two projects share an acronym only.

Talking with us

The most useful conversation at this stage is a 30-to-45-minute exploratory call with an SDO’s policy lead or general counsel, supplemented by a standards-development volunteer who can speak to how the SDO’s content is actually structured. The goal is not to sell a deployment. The goal is to understand whether CLASP’s architecture maps cleanly to the SDO’s existing license registry, editorial pipeline, and risk posture, and to identify the smallest pilot that would let the SDO learn something useful.

Reach Andy Robinson at arobinson@consynsys.com. The project is in its early phase and feedback at this stage carries proportionally more weight than feedback later; the specification is still open to revision in places where SDO requirements surface real gaps.