AI agents can read, buy and sign in on people’s behalf, but websites may be left carrying the costs and risks. Cloudflare is proposing rules for keeping user and site responsibilities clear, while offering customers several controls already.
Cloudflare Watch analysis
What happened
Cloudflare’s position paper argues that agent traffic can leave websites exposed to disputed purchases, credential abuse and content being used without a visit or advertising impression. The paper proposes an “Intermediation Principle”: the relationship between a user and a website, including its security duties, should persist when software acts between them. PPC Land reports that the paper was written by Bryan Becker and sets out 14 problems for websites.
The proposals include a four-level scale for user identity, Web Bot Auth for identifying agents, and a way to state what agents are permitted to do. The paper also backs a rate-limiting approach called MOLE and OAuth for account access. PPC Land says three related Cloudflare controls are already live for customers: analytics for personal-assistant traffic, per-agent access policies that can block agents on ad-supported pages, and Account Abuse Protection in early access.
Why it matters
The argument is about who pays when an agent browses or acts on a person’s behalf. If a site cannot tell whether a purchase was authorised, or an automated login is legitimate, the website may inherit costs even though the agent is serving someone else. Clearer identity and permissions could help, but standards and adoption matter as much as the proposal.
Our read
This is a useful attempt to make agent access a question of accountability, not just traffic volume. Cloudflare has a commercial interest in bot management, and PPC Land says the paper offers no traffic, dispute or adoption figures. That does not make the proposal irrelevant; it does mean the case needs evidence beyond the company’s own diagnosis. “The agent did it” is not a particularly satisfying answer for the site left holding the bill.
What to watch
- Whether Cloudflare’s proposed identity and permissions approaches gain support from standards bodies.
- Whether the company publishes evidence about agent traffic, disputes or the costs it describes.
- How widely customers adopt the three controls, and what practical results they report.
Discussion spark: If an AI agent causes a disputed purchase or abuses a login, should responsibility sit mainly with the agent provider, the user, or the website?
Sources and evidence
- Cloudflare says websites pay for AI agents as Meta's Muse learns to talk – PPC Land (9 November 2026, 11:38 UTC)
Independent WittyWires tracker for public updates about Cloudflare. Not affiliated with or endorsed by Cloudflare; this is not an official account.