TL;DR: On October 20, 2026, W3C VP Technical Strategy Philippe Le Hegaret finalized "W3C and the State of the Web in the AI Era," a Strategy Team discussion paper proposing nine principles for how AI agents should behave on the web, plus a new Agentic Web Working Group and work items for more than a dozen existing groups. It isn't a standard — it's a discussion paper meant to set the agenda at TPAC 2026, where the Web and AI Interest Group will take it up on October 29. But it's the clearest signal yet that the organization behind HTML, CSS and the rest of the platform thinks the current free-for-all around AI agents browsing on people's behalf can't continue unaddressed.
The paper's framing is blunt about what's missing. As Le Hegaret puts it, "the implicit contract of content providers with search engines — crawl my content, send me visitors — has yet no systematic equivalent for AI systems." That gap is exactly where teams doing real getting consent, security, and transparency right when you build AI agent integrations into a product keep running into the same unanswered questions: what is an agent allowed to do on a user's behalf, how does it prove it, and who is liable when it gets something wrong.
The declaration is short on legal language and long on principle, which is deliberate — this is a starting point for standards work, not a finished spec. Per reporting on the paper, the nine points establish that the web serves all people, including those using AI agents; that an agent acting for a person functions as a user agent owing that person loyalty, transparency and safety; that people retain real, informed control over consequential actions rather than standing consent; that the web stays one interoperable system without agent-only lock-in channels; that existing web security models apply to agents the same way they apply to scripts; that creators get a voice in how their work is used through fair, machine-readable agreements; that human accessibility outranks machine optimization; that trust requires verifiable provenance and credentials; and that new agent-facing features must be necessary, efficient and transparent about the resources they consume.
The paper's own framing matters here: a 2024 W3C Team report had already flagged the risks and opportunities of machine learning on the web, but the shift it describes in 2026 is qualitatively different — AI agents have moved from reading pages to interacting with them, and every major AI platform now embeds an agent-driven browser. That's the practical reason crawling, consent, payments, identity and accessibility all needed to be addressed together rather than piecemeal: an agent that fills out a form, completes a purchase, or authenticates as a user touches all five at once, and today each of those surfaces has its own ad hoc rules, if it has any at all.
Nothing here is a ratified standard, and nothing changes in a browser or an API this week. But if you're building or specifying anything that lets an AI agent act on a user's behalf — booking something, moving money, editing a profile, calling an internal API — the nine points are a reasonable checklist to hold your own design against today, before anyone forces you to. Concretely: can you point to where a human gave real, specific consent for the consequential action the agent just took, not a blanket "yes" buried in onboarding? Does the agent identify itself distinctly from a human user in your logs and rate limits, the way the paper's security-boundary point implies it should? And if a creator's content fed the agent's output, is there any machine-readable signal about how that's allowed to be used? Teams skip these questions because no regulator or spec currently requires answering them — which is precisely the gap this paper is trying to close before it becomes someone's incident report.
Not soon, and the paper is explicit that nothing in the declaration is a standard yet. The near-term calendar: the Web and AI Interest Group takes up the paper at TPAC 2026 on October 29, feedback is being collected through the public [email protected] mailing list and the w3c/webai-roadmap GitHub repository, and the proposed Agentic Web Working Group — whose draft charter already covers WebMCP, agent identity, permissions and transparency — still needs to be chartered before it can produce an actual Working Draft. Realistically, treat this as the opening move in a multi-year standards process, the same shape WCAG, Web Components and most other W3C work has taken, rather than something with a near-term ship date.
No. It's a discussion paper from the W3C Strategy Team, finalized October 20, 2026, meant to set the agenda for standards work — not a Recommendation, Candidate Standard, or anything with normative force.
A new W3C working group proposed in the paper, with a draft charter covering WebMCP, agent identity, permissions and transparency for AI agents acting on the web.
Philippe Le Hegaret, W3C's VP Technical Strategy, prepared it on behalf of the W3C Strategy Team.
At TPAC 2026, in the Web and AI Interest Group session on October 29, 2026. Public feedback is being collected via the public-webai mailing list and the w3c/webai-roadmap GitHub repository beforehand.
The web serves everyone, including people using agents; agents owe the people they act for loyalty, transparency and safety; people keep real control and consent over consequential actions; the web stays one interoperable system; existing security models apply to agents; creators get a voice via machine-readable agreements; human accessibility outranks machine optimization; trust needs verifiable provenance; and new agent features must be necessary, efficient and resource-transparent.
Primary source: W3C's "W3C and the State of the Web in the AI Era" announcement. Independent coverage confirming the nine points and the Agentic Web Working Group proposal: Heise's report and ppc.land's breakdown.