# Le Monde evaluates AI-agent access for subscribers > Source: > Published: 2026-06-17 04:53:49.618732+00:00 Photo: digiday.com · rights & takedowns Digiday reports that French publisher Le Monde is blocking nearly all non-human traffic unless a licensing deal exists, including Googles AI training crawler Google Extended. At Fastlys Xcelerate event in London, Le Monde CTO Paul Laleu told Digiday the publisher is "figuring out" how to recognize paying readers who access articles via AI agents rather than the homepage or app. Laleu cited early technical standards  model "apps" and OAuth-style login flows  as potential plumbing for an agent to assert a user's subscriber status, per Digiday. Fastly data cited by Digiday shows AI-driven traffic growing at 6.5 times the rate of human traffic. Digiday did not report a finalized approach or public statement from Le Monde beyond Laleu's remarks. What happened Digiday reports that Le Monde now blocks almost all non-human traffic unless there is a licensing deal in place, including Google's AI training crawler Google Extended . At Fastly's Xcelerate event in London, Le Monde CTO Paul Laleu told Digiday the publisher is "figuring out" how to maintain its subscription relationship when readers arrive via AI agents instead of directly through the homepage or app. Per Digiday, Laleu described early technical standards  model apps and OAuth-style authentication extensions  as potential mechanisms for an agent to assert "this user is logged in and already pays" so the publisher can apply paywall rules. Fastly's data cited by Digiday indicates AI-driven traffic is growing at 6.5 times the rate of human traffic. Technical details Editorial analysis - technical context: Industry discussion is focusing on two families of primitives. One is agent-to-tool integration, commonly described as model apps or connectors, which let models query external services in a controlled way. The other is delegated authentication patterns, analogous to OAuth, that would let an agent present a delegated credential proving end-user entitlements without exposing passwords. On-device assistants (examples cited by Laleu include Apples recent updates and Googles Gemini ) increase the chance that requests will originate from software intermediaries rather than browsers, which changes how publishers need to verify identity and entitlement. Context and significance Editorial analysis: Publishers have long relied on browser- and app-based signals for paywalls and analytics. The rise of AI-mediated access separates the user agent from the original UI, creating friction for subscriber recognition, consent flows, and analytics. The technical approaches Laleu mentions are familiar patterns in modern web auth, but applying them to agent-mediated requests raises questions about privacy-preserving entitlements, replay protection, and how licensing contracts map to automated requests. Fastly's growth metric cited by Digiday underscores why publishers are prioritizing this problem: agent traffic is expanding faster than human traffic, amplifying potential revenue and attribution issues. What to watch For practitioners: watch for specification work and experiments from browser vendors, assistant platforms, and standards bodies that define delegated-auth flows for agents. Also monitor pilot implementations from publishers and CDN telemetry (like Fastly's) to see how agent headers, rate-limiting, and entitlement tokens evolve. Observables that will matter include: adoption of token-exchange patterns for agents, standard claims that express subscriber status, and how CDNs and paywall providers validate delegated credentials. Scoring Rationale This is a notable operational story for publishers and platform engineers: agent-mediated access changes authentication and paywall enforcement. It is not a frontier-model release, but it affects production infrastructure, access control, and monetization. More AI Agents news → Practice interview problems based on real data 1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with. Try 250 free problems