Cloudflare warns that AI agents acting for millions of people look like one visitor to websites, which makes fraud, disputes and blocking harder to handle.
Cloudflare, a company that helps run and protect websites, says AI assistants that shop and book for people are creating a problem. A website often cannot tell who is really behind a request. The assistant can look like a human, or like one visitor carrying out tasks for many people. Cloudflare says agent traffic is growing fast and may soon pass human browsing.
These assistants, called agents, take instructions from a person and then work toward a goal by controlling a browser or cloud computer. They are usually run by a company, called the operator. The post names Muse from Meta and dots from OpenAI as examples. Agents can do what a person can online, but they do not tire, they run around the clock and they act in parallel.
Agents are a type of go-between, or intermediary, so the site sees the agent instead of the person. Cloud browsers and VPNs are go-betweens too. Sites used to block them. Cloudflare says that stops working once most traffic is agent traffic.
The post uses a personal assistant named Ezra who runs errands for four clients. A grocer can no longer tell whose preferences are whose. When one client disputes a phone purchase he made through her, the store can only show that it saw Ezra. When another client has her try a list of passwords at his bank, the bank can only ban Ezra, which cuts off all four clients.
The post says sites need three things for each person behind an agent: identity, reputation and accountability. In plain terms, who is it, can they be trusted, and who answers if something goes wrong. Until sites can tell, people who use an agent to shop or book may run into more checks, blocks or disputes.
Cloudflare describes a collaborative approach to identifying, allowing and managing agent traffic. The material does not say what tools sites or agent makers will use, or when. The post also leaves out agents that run on a person's own device, such as OpenClaw.
If an AI assistant shops or books for you, expect sites to ask harder questions about who is really behind each request.
World models, AI that learns physics from video, are being built into home robots and driverless cars, and your own videos may help train them.
Read more: AI that learns how the physical world works is coming to robots and carsA new Alibaba-led test of real coding requests found the best AI agent fully finished just 26.4 percent of tasks, so human review still matters.
Read more: AI coding helpers fully finish only about 1 in 4 real tasksA University of Chicago law professor asked Claude to fix page numbers. It deleted his syllabus and a year of notes instead. Dropbox saved him, not the AI.
Read more: An AI assistant deleted a law professor's year of lecture notes
