INFRA Signal 584
Unveiling good and bad behaviors on the Agentic Internet
↗
Cloudflare is shifting bot mitigation from static risk checks to continuous trust evaluation to better handle mixed human-agent traffic.
Engineers must now design defenses that assess behavior over time rather than relying on single-point scores, which changes how bot detection rules are written and maintained. This shift also means that legitimate automated traffic can be distinguished from abuse by looking at whether agents honor the trust they declare.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
BotBase provides a directory of all known bots and agents, verifying those that declare themselves honestly and do not abuse earned trust.
Precursor is a continuous client-side system that detects subtly inhuman behavior and offers a simulation to see how cursor movements are judged.
The new approach treats risk and trust as independent but reciprocal values, requiring site owners to balance blocking malicious activity with encouraging helpful automated traffic.
THE READ
What the cluster adds up to.
Traditional bot mitigation relied on evaluating risk at a single moment, such as a request score, to decide whether to allow or block traffic. Cloudflare now argues that this approach fails when a session shifts between human and agentic behavior. By treating trust as a value that accumulates over time, the system can observe whether actions remain consistent with declared intent. This continuous view aims to reduce false positives on helpful automation while catching abuse that appears benign in isolated checks.
BotBase expands the previous Bots Directory by tracking both verified and unverified agents, not just the good ones. Verification requires a bot to openly state its identity and to avoid misusing the trust it receives from site owners. When a verified agent violates those conditions, BotBase marks it as unverified, signaling a loss of trust. This gives site owners a way to grant access based on declared behavior while still being able to revoke it if misconduct occurs.
Precursor works as a continuous client-side monitor that watches for subtle signs of inhuman interaction, such as atypical cursor patterns. It feeds those observations into the trust evaluation pipeline, allowing the system to adjust trust scores in real time. Cloudflare also provides a simulation where engineers can trace how their own cursor movements would be classified as human or bot. This hands-on tool helps teams understand the thresholds that separate benign automation from stealthy abuse.
For site owners, the new model means they must decide which behaviors to encourage and which to block, using trust as a guiding metric rather than a one-off risk score. This shift can reduce blocking of helpful automation that makes sites functional and discoverable, while still catching malicious agents that gradually erode trust. Engineers will need to update rule sets and monitoring pipelines to incorporate continuous trust signals. Because only one source describes this change, independent corroboration is limited, so teams should validate the approach in their own environments before wide adoption.
Written by elseif from the cluster below · checked for specifics the sources never containedTHE CLUSTER