Cloudflare changes AI Security categories: review WAF rules
Review category-based WAF conditions after the AI Security for Apps update, with a practical test plan for teams running an AI assistant.
On October 7, 2026, Cloudflare announced changes to AI Security for Apps categories. Rules referencing outdated names may remain valid while no longer matching the intended category. Cloudflare announcement.
Who should review their setup
Advertisers and publishers already using this service to protect an AI assistant should involve their technical team. The documentation exposes categories through cf.llm.prompt.unsafe_topic_categories. Detection documentation.
Our editorial takeaway: a rule that saves successfully still needs a behaviour check. A team might see no configuration error and assume its intended control remains effective. The announcement alone does not establish a fault on any particular website.
A practical verification plan
Inventory rules using the field, assign an owner and preserve the current expressions. Compare their values with the documentation. Do not infer a mapping between old and new categories merely because their names look similar.
In a test environment, run ordinary permitted requests alongside safe control examples prepared by the security owner. Record the expected action, observed result, language and rule identifier. Include everyday registration and support questions: unnecessary blocking can also disrupt customer service.
Repeat the same checks after an approved change. Completion means reproducible results and a documented rollback route, rather than an empty error panel. Keep customer personal data out of test prompts.
These are editorial recommendations, not an announcement of an AffiliateCore integration. Prompt filtering and affiliate traffic-quality assessment remain separate checks.