Unveiling good and bad behaviors on the Agentic Internet

Cloudflare is moving bot mitigation away from a single point-in-time Risk score toward continuous Trust evaluation. The company frames this around good and bad behaviors from bots and agents on what it calls the Agentic Internet, and describes how those behaviors are scored by systems including BotBase and Precursor.

Under the new model, traffic is not judged only at request time. Precursor looks at interaction signals such as cursor movement, while BotBase contributes to classifying bot and agent behavior as good or bad rather than treating every automated client the same. Cloudflare also offers a Precursor Trace simulation so operators can see how their own cursor movements would be labeled as human or bot.

For engineers shipping agent-driven clients, scrapers, or automation, the shift matters because a one-shot risk check no longer fully describes how traffic will be treated. Continuous Trust evaluation means session-long behavior, including mouse and pointer patterns, can change whether a path is allowed, challenged, or blocked even when headers and fingerprints look legitimate.

That sits against a market where agents, headless browsers, and legitimate automation all hit the same edges as abusive bots. Cloudflare’s good-versus-bad behavior framing is an attempt to separate useful agents from harmful ones without relying only on static blocklists or a single request-level risk label.

The practical takeaway is to test real client interaction paths against Precursor Trace and review how BotBase and continuous Trust evaluation would classify them, then adjust automation so cursor and session behavior match the trust profile you need. Watch how Cloudflare exposes and tunes good and bad behavior signals for agent traffic as more production clients go through this model.