Cloudflare has launched the first part of Adaptive Intelligence, a bot-detection engine designed to change as attackers change their tactics. Its central machine-learning model now retrains continuously on live network traffic rather than remaining fixed until a scheduled release.
The company says static rules give bot operators a stable target: attackers can test proxies, fingerprints and request patterns until they discover what passes. Adaptive Intelligence instead combines signals across requests and time windows, looking for behavior shared across addresses, clients and sessions. Cloudflare processes more than a trillion requests a day, giving the system a broad view of new automation techniques.
Two additional components are planned but are not part of the initial launch. One would generate temporary rules that appear and disappear before attackers can reliably reverse-engineer them. Another would use corrections from customers and Cloudflare’s own measurements as training signals.
The engine sits behind Cloudflare’s existing bot score, alongside behavioral checks, JavaScript fingerprinting and recognition of verified crawlers. Customers should therefore see an evolving score rather than a separate product interface. Cloudflare’s performance and false-positive claims still need to be tested in real deployments as the later components arrive.