Unveiling good and bad behaviors on the Agentic Internet

Cloudflare is shifting bot mitigation from point-in-time Risk assessment to continuous Trust evaluation. The change targets the Agentic Internet, where traffic mixes human browsers, traditional bots, and autonomous agents. Cloudflare systems assess both good and bad behaviors from bots and agents instead of scoring a single request in isolation.

On the technical side, assessment runs through BotBase and Precursor. BotBase supports the broader bot and agent classification stack. Precursor focuses on behavioral signals such as cursor movement patterns. Cloudflare also offers a Precursor Trace simulation so builders can run their own cursor movements through the same logic and see whether the system labels the session human or bot.

For engineers and builders, the shift changes how you instrument and test agent traffic. Point-in-time Risk checks fit simple scrapers and one-shot API calls. Continuous Trust evaluation expects multi-step agent sessions, tool use, and long-lived browser control. If you ship agents that drive UIs, scrape, or automate workflows behind Cloudflare, you need to understand which behaviors look trusted and which look abusive.

The market context is the rise of agentic clients that are neither clean human sessions nor classic bot farms. Operators need to allow useful automation while blocking harmful automation. Cloudflare is positioning BotBase and Precursor as the machinery for that split: continuous Trust signals instead of a single Risk score at the edge.

Practical takeaway: try the Precursor Trace simulation against real cursor paths from your product’s automation and from genuine users. Compare how trusted agent flows and known-bad bot patterns score under continuous Trust evaluation. Watch how BotBase and Precursor classify your agent’s multi-step sessions, and adjust agent pacing, interaction patterns, and access paths before you rely on those agents in production behind Cloudflare.