// AI Platform Drift
Bot Management ML model switches from fixed versions to continuous retraining
Cloudflare developer changelog · published 2026-10-16 04:41 UTC
WHAT CHANGEDCloudflare's Bot Management machine learning model, which generates bot scores, changed from shipping numbered fixed versions to continuous retraining under a system called Adaptive Intelligence. New model versions now run in shadow mode for validation before gradual rollout. Enterprise customers with Auto-update machine learning enabled receive this automatically as of 31 August 2026.
WHY IT MATTERSBot score behavior will change without version number updates, making it impossible to pin to a specific model version. If your application logic depends on consistent bot scoring thresholds or you need reproducible results for testing, the continuous retraining removes the ability to control when scoring behavior changes.
WHAT TO DOCheck your Bot Management configuration dashboard to confirm whether Auto-update machine learning is enabled. If your application has hardcoded bot score thresholds or relies on stable scoring for A/B tests or compliance validation, document current behavior now and establish monitoring for score distribution changes, since version pinning is no longer an option.
Model-reported confidence 85% (not measured accuracy)
Primary source ↗
written by anthropic/claude-sonnet-4.5
Show the reasoning — raw diff, triage verdict, confidence
RAW DIFF
+## Bots - Bot Management's machine learning model now retrains more frequently instead of shipping fixed versions +https://developers.cloudflare.com/changelog/post/2026-08-31-adaptive-intelligence/ +Mon, 31 Aug 2026 00:00:00 GMT + +We have launched Adaptive Intelligence, a new bot detection engine built to close the gap between when an attack pattern first appears and when protection against it reaches every customer. You can read the announcement blog ↗︎ for the full story behind why we built it. +Adaptive Intelligence, the machine learning model behind your bot score, now retrains more frequently instead of shipping as a fixed, numbered version. As new bypass tools and bot frameworks appear, the model learns from them and folds that knowledge in without waiting for a scheduled release. +Every update is validated before it affects you. New model versions run in shadow mode first, scored but not acted on, and are compared against the current model before rolling out gradually. +Enterprise customers with Auto-update machine learning switched on under Bot Management configurations get Adaptive Intelligence automatically, with no version to migrate and nothing else to configure. +To learn more, refer to Adaptive Intelligence and Machine Learning models.
HOW IT WAS JUDGED
Triage signals below are model-reported assertions, preserved verbatim; they are not independent arithmetic verification.
| triage verdict | MATERIAL at 0.92, needed 0.5 |
|---|---|
| triage reasoning | Bot detection model behavior fundamentally changes from versioned releases to continuous adaptive retraining, affecting how customers receive protection updates. |
| signals | new feature launch with behavioral change · model retraining frequency increased (fixed versions → continuous) · shadow-mode validation process introduced · automatic rollout for Enterprise customers · configuration impact for Bot Management users |
| suppression | — |
| triage model | anthropic/claude-haiku-4.5 ($0.00000) |
| writer model | anthropic/claude-sonnet-4.5 ($0.00000) |