Muse is the web’s busiest agent — and websites are footing the bill
Cloudflare’s November 5 paper says sites bear the primary cost of AI-agent visits in nearly every case it lists, and Muse — at 72% of AI agent requests — is the biggest visitor. It proposes identity and rate-limiting standards, with some controls already live.
Paper by Bryan Becker, “Agents aren’t humans (but they kinda are)”, published November 5.
Lists 14 problems AI agents create for websites; “in nearly every challenge above, the site bears the primary cost.”
Examples: shops that can’t tell if a purchase was authorised, login forms hammered by software holding real credentials, articles summarised by a bot that never loads an ad.
Muse accounts for an average of 72% of daily AI-agent requests, 5.2x the volume of ChatGPT Agent.
Arkose Labs found 97.7% of legitimate AI agent sessions carry no verified identity.
DataDome counted 29.1 million Muse requests on its network between September 22 and October 3.
Proposes a four-level scale of user identity: nothing, session, account, unique.
Favours Web Bot Auth for agent identity and a rate-limiting scheme called MOLE that works without user identifiers.
Calls for a machine-readable way to state what agents may do, since robots.txt wasn’t built for the job.
A coalition called Trusted Agent Delegation will take the problems to standards bodies.
Analytics for personal assistant traffic is live.
Per-agent access policies can block agents only on pages carrying ads.
Account Abuse Protection is in early access.
Cloudflare published no traffic, dispute, or adoption figures for the new controls.