Skip to content

Machine Learning models

Last updated View as MarkdownAgent setup

Enable auto-updates to the Machine Learning models

Enterprise customers with auto-updates to the machine learning model turned on get Adaptive Intelligence. To check or change the setting:

  1. In the Cloudflare dashboard, go to the Security Settings page.

    Go to Settings ↗
  2. Filter by Bot traffic.

  3. Go to Bot Management.

  4. Under Configurations, select the edit icon for Auto-update machine learning and turn it on.

What will change

Enabling auto-updates moves your traffic onto the model behind Adaptive Intelligence. This model retrains more frequently than fixed, numbered versions were released, so your bot score reflects newly identified attack patterns as the model learns them, rather than only when a new version ships.

If you are on an older, fixed Machine Learning model, you will see a score change to requests scored by the Machine Learning source as soon as you enable auto-updates.

Risks of not updating

If you do not enable auto-updates, your traffic continues to be scored by your current fixed model version. Cloudflare does not maintain or monitor fixed model versions after they are superseded, and scoring accuracy degrades as attack patterns evolve beyond what that version was trained on.

Model versions and release notes

The following table lists released versions through v9, the final fixed version. Enabling auto-updates moves you off numbered versions entirely, onto Adaptive Intelligence.

Version Release Notes Launch Date
v1 First Machine Learning Model released. Q1 2019
v2 Introduced dynamic inter-request features to leverage the Cloudflare network to detect new bots more accurately.

Feedback other Bot Management detection mechanisms to the machine learning model to more accurately detect bots.
Q1 2020
v3 Fixed accuracy issues under some conditions in the previous version. Q2 2020
v4 Improved scoring for iOS devices.

Fixed scoring inaccuracy in Firefox builds.
Q1 2021
v5 Recalibrated model for the removal of _cfduid cookie ↗︎.

Introduced new signals to reduce false negatives.
Q2 2021
v6 Significantly improved scoring for native Android application traffic.

Improved scoring on the newest versions of Chromium browsers.
Q1 2022
v7 Increased recognition of distributed botnets.

Improved HTTP/3 scoring.
Q1 2024
v8 Improved detection of residential proxies.

Increased weight on network level traffic characteristics.
Q2 2024
v9 Improved model consistency and model efficacy against randomization attack techniques Q2 2025

Was this helpful?