Technology

Cloudflare says agents break website trust

Cloudflare warns remotely hosted AI agents hide users from sites, breaking fraud controls and accountability.

StartupHub.ai Staff

By StartupHub.ai Staff, reviewed by Daniel Singer, Editor

2 min read

AI agent acting for multiple users in front of a website
Credit: Cloudflare

Cloudflare says agent traffic is growing fast and may soon eclipse human browsing, leaving sites unable to recognize or charge the real user. The warning covers remotely hosted agents that act as intermediaries, no local access needed.

Bryan Becker, writing on the Cloudflare blog, centers the argument on Ezra, a hypothetical personal assistant who shops for four clients.

Amy is her best client, a local CEO, plus Bob, Charlie and Debra. The corner grocer used to know Amy and what she liked. Now he only sees Ezra carrying orders for four people, and he does not want to turn her away because she brings volume.

Then Bob disputes the iPhone charge.

Becker says Bob asks Ezra to buy the phone, then tells the store he never authorized it and keeps the device. The store has to prove Bob asked for it, but it only ever saw Ezra. In the next case, Charlie hands Ezra a list of passwords to try at his bank. The bank sees what looks like an attack, yet banning Ezra would also ban Amy, Bob and Debra.

That is the intermediation problem, Becker argues. Cloudflare proposes what it calls the Intermediation Principle, that the business relationship between users and sites must hold regardless of intermediation. Sites still need to answer three questions, he writes: is the person truthful about who they are, is that person trustworthy, and who answers for the action.

The push for agents is already public. OpenAI describes its dot as an always-on agent in ChatGPT that keeps making progress between conversations, according to OpenAI help center. Fortune reported OpenAI pitched Dots at its San Francisco developer event a month ago and then took the roadshow to London. Meta is moving the same way with Muse, which Becker names alongside dots, while local tools like OpenClaw are explicitly out of scope for his post.

That scope limit matters. Becker is talking about agents hosted somewhere other than the user device, not a browser on your laptop. His accountability model assumes an operator to identify, which does not exist when the agent runs locally.

Who proves Bob told Ezra to buy the phone when the store never saw Bob?

Sources

Get the free daily brief every morning

Every morning the day's AI news, new AI startups and new funding rounds. Free and takes a few minutes to read, unsubscribe at any time.

Join 64,000 other subscribers.

TopicsCloudflareOpenAIMetaDotsMuseChatGPTOpenClawBryan Becker

StartupHub.ai Staff

Written by

StartupHub.ai Staff

Editorial team

The staff writers of StartupHub.ai, ranging from investment analysts to avid AI tool users, early adopters and critical enthusiasts. Backgrounds span engineering, business and the arts. We hold every piece to rigorous standards of research and review.

Daniel Singer

Written by

Daniel Singer

Editor, reviewed this article

Daniel Singer is an AI expert who has been writing about, covering and building AI technology since 2018.

Test your Agent Readiness now

Paste your website. How well ChatGPT, Claude and the other agents can read and use a site, in about a minute.

Watch

The videos behind the news, one flip at a time

The feed mixes the videos from our articles with startup demos and you flip through them like TV channels for free with no signup.