Cloudflare lets sites set access rules for individual AI assistants

Cloudflare introduced analytics and individual access controls for personal AI agents on November 4. Website owners can identify supported assistants visiting on a person’s behalf and make exceptions to their broader agent policy. The change covers software that retrieves information or browses sites for users, including shopping and support tasks.

Owners can allow an agent, block it, or block it on pages with ads. A new Personal Assistants category and per-agent request counts show up in the analytics interface. Selecting an assistant narrows the view to its requested paths, response codes and source networks, according to Cloudflare.

The underlying policies separate three behaviors. Search collects or indexes content for later answers; Agent acts in real time for a person; Training collects material for model training or fine-tuning. A crawler used for both search and training also falls under Training. Policies cover verified bots with the relevant behavior and matching unverified bots.

Personal agents initially inherit the overall Agents policy. Changing that policy updates their settings, after which owners can choose individual actions again. Existing Search and Training settings remain in place. Advertising-page restrictions use Cloudflare’s detection of ads. The Allow action means the AI access policy does not block matching requests.

Robots.txt communicates a separate layer of preferences. Crawler operators decide whether to follow its instructions. Cloudflare’s Bot Preference Sync adds managed directives when restrictive policies are configured. Where a site already serves the file, the service prepends its managed section and preserves the existing content in the combined response.

The figures also have a measurement boundary. Security Analytics calculates top statistics from sampled requests, with the sampling rate available beside each statistic. The dashboard includes traffic that passed normally as well as requests affected by security products. Domain-level analytics are included across plans, while retention, query windows, statistics and filters vary.

Where assistants can be linked to an account

Identifying the account behind a login requires Account Abuse Protection, which is in early access for Bot Management Enterprise customers. Access is free for a limited period before general availability. Sites must enable User ID and specify where their login or sign-up requests carry the account identifier; Cloudflare turns that value into a per-zone hashed identifier.

Recognized assistant login and sign-up events enter that account’s history. Teams can inspect associated assistants and event counts, then target a suspicious account individually. Unrecognized assistants appear as ordinary traffic. This account-linked view covers login and sign-up events, and other Cloudflare customers cannot access those user profiles.

Cloudflare’s verification criteria require honest identification through a cryptographic signature, published IP addresses with a stable user-agent, or reverse DNS. They also require non-abusive behavior, including respecting crawl instructions and reasonable request rates. Its classification distinguishes a narrowly operated bot from an intermediary service used by many people: verifying that service identifies its operator, while the end user remains a separate party.

Cloudflare describes preserving customer relationships, subscriptions and user authorization through personal agents as further work. The announcement gives neither a concrete list of assistants eligible for individual controls nor a release date for those subsequent capabilities.