Cloudflare has introduced Adaptive Intelligence, a new bot-detection engine that sits behind its Bot Score. The company describes it as a shift from a wall-centric approach to one that makes bypass efforts slow and costly for attackers.
The goal is to reverse the economic incentives of bot operators by ensuring that attempts are harder to justify economically, even as attackers experiment with increasingly sophisticated configurations.
WHAT ADAPTIVE INTELLIGENCE IS MEANT TO DO
Adaptive Intelligence is designed to change defenses continuously in response to evolving bot tactics. Cloudflare characterizes it as a non-deterministic, signal-rich approach that reduces the predictability attackers rely on when probing defenses. TechStaged has also covered New arXiv preprint introduces FedEHR-Agents for Federated EHR Modeling.
The system aggregates multiple signals across Cloudflare’s network to assess the likelihood of automated abuse for every request, rather than relying on a fixed set of rules.
HOW THE ENGINE OPERATES IN PRACTICE
The engine follows a loop: observe signals, train on live traffic, deploy new model weights across the network, and validate changes before they go live.
Key components highlighted include continuous retraining on live traffic, disposable rules that appear and disappear to inject noise for attackers, and cross-site learning from traffic patterns to adapt to new attack techniques.
- Observe signals from across the Cloudflare network (including JA4 TLS fingerprints, request structures, challenge outcomes, session behavior, and network reputation) and client-side telemetry from Turnstile and Precursor
- Train the model continuously on live traffic so updates reflect current threats
- Deploy model weights across the network automatically, with validation in shadow mode before affecting real visitors
- Validate new versions by comparing signals like challenge solve rates to ensure real users are not adversely impacted
WHAT THIS MEANS FOR DEFENDERS AND ATTACKERS
Cloudflare argues that attackers are constrained mainly by time and their proxy networks, while defenders must manage accuracy to avoid impacting legitimate users. By contrast, Adaptive Intelligence aims to ensure that any single signal is not sufficient for a bidirectional, fixed rule, thereby increasing the cost and effort required for bot operators.
WHAT COMES NEXT
The first component of Adaptive Intelligence centers on machine learning at the core of the bot score, with continuous retraining. Cloudflare notes that two additional components—disposable rule generation and learning from protected traffic—will follow as the system expands across networks.
The approach emphasizes memory of past attacks while avoiding fixed targets, so that shifting tactics remain challenging for attackers over time.
RELATED COVERAGE
SOURCES
- Cloudflare Blog: Introducing Adaptive Intelligence: undermining the economics of every bot attack Published · Primary source
