See the agents, set the terms: new visibility and control for personal agents

A personal agent is an AI assistant that browses and acts on websites for one person, comparing products, finding support pages or completing a purchase. To a site, those visits look like any other automated traffic, which leaves owners choosing between blocking agents that may bring customers and admitting traffic they cannot identify. Cloudflare now labels that traffic by agent, using its directory of identified bots, and lets sites set rules for each supported agent. The post does not list which agents are supported.

The visibility comes in two places. In Security Analytics, Personal Assistants is a new category beside verified bots, likely humans and automated traffic, and a widget lists which agents visit and how many requests each sends; filtering on one agent shows the paths it requests, the status codes it gets and the networks it comes from, so an owner can see whether it reaches checkout or keeps hitting errors. Sites using Account Abuse Protection, which is in early access, can also see which user account each agent acts for, whether a login or sign-up came from an agent, and which agents an account has used before. That lets a fraud analyst tell an assistant the customer has used for weeks from one that appeared today, and block a suspicious account through a firewall rule on its hashed user ID without blocking the agent for everyone else.

The control is an extension of the AI bot policies Cloudflare introduced this summer, now renamed AI Crawler and Agent access policies. An owner sets a default for agents and can override it per supported agent with Allow, Block on pages with ads, or Block, while existing search and training crawler settings stay unchanged. The ad option is aimed at sites paid for by advertising: when a new site joins, Cloudflare asks whether it monetizes pages with ads and tailors its recommended policies to the answer, and its example is a publisher that blocks an agent on pages where ads support the content while another allows the same agent everywhere so readers can find its articles. Cloudflare is open about what this does not solve. Recognizing the agent does not tell a site about the person behind it, such as whether they are an existing customer or subscriber or what they authorized the agent to do. Cloudflare calls preserving that context its next goal and says reaching it will require privacy-aware ways to carry user context through an agent, help sites assess reputation and establish accountability when something goes wrong.