CVE-2021-37678: Arbitrary code execution due to YAML deserialization

Published Aug 12, 2021
·
Updated

Impact TensorFlow and Keras can be tricked to perform arbitrary code execution when deserializing a Keras model from YAML format.

python from tensorflow.keras import models

payload = ''' !!python/object/new:type args: ['z', !!python/tuple [], {'extend': !!python/name:exec }] listitems: "import('os').system('cat /etc/passwd')" ''' models.modelfromyaml(payload) The implementation uses yaml.unsafeload which can perform arbitrary code execution on the input.

Patches Given that YAML format support requires a significant amount of work, we have removed it for now.

We have patched the issue in GitHub commit 23d6383eb6c14084a8fc3bdf164043b974818012.

The fix will be included in TensorFlow 2.6.0. We will also cherrypick this commit on TensorFlow 2.5.1, TensorFlow 2.4.3, and TensorFlow 2.3.4, as these are also affected and still in supported range.

For more information Please consult our security guide for more information regarding the security model and how to contact us with issues and questions.

Attribution This vulnerability has been reported by Arjun Shibu.

Other sources

TensorFlow is an end-to-end open source platform for machine learning. In affected versions TensorFlow and Keras can be tricked to perform arbitrary code execution when deserializing a Keras model from YAML format. The implementation uses yaml.unsafeload which can perform arbitrary code execution on the input. Given that YAML format support requires a significant amount of work, we have removed it for now. We have patched the issue in GitHub commit 23d6383eb6c14084a8fc3bdf164043b974818012. The fix will be included in TensorFlow 2.6.0. We will also cherrypick this commit on TensorFlow 2.5.1, TensorFlow 2.4.3, and TensorFlow 2.3.4, as these are also affected and still in supported range.

Affected Software

15 affected componentsFixes available
pip/tensorflow-gpu=2.5.0
2.5.1
pip/tensorflow-gpu>=2.4.0<2.4.3
2.4.3
pip/tensorflow-gpu<2.3.4
2.3.4
pip/tensorflow-cpu=2.5.0
2.5.1
pip/tensorflow-cpu>=2.4.0<2.4.3
2.4.3
pip/tensorflow-cpu<2.3.4
2.3.4
pip/tensorflow=2.5.0
2.5.1
pip/tensorflow>=2.4.0<2.4.3
2.4.3
pip/tensorflow<2.3.4
2.3.4
Google TensorFlow>=2.3.0<2.3.4
Google TensorFlow>=2.4.0<2.4.3
Google TensorFlow=2.5.0
Google TensorFlow=2.6.0-rc0
Google TensorFlow=2.6.0-rc1
Google TensorFlow=2.6.0-rc2

Remediation

Recommended actions to resolve this vulnerability, in priority order.

  1. Upgrade

    Upgrade pip/tensorflow-gpu to a version that resolves this vulnerability.

    Fixed in 2.5.1
  2. Upgrade

    Upgrade pip/tensorflow-gpu to a version that resolves this vulnerability.

    Fixed in 2.4.3
  3. Upgrade

    Upgrade pip/tensorflow-gpu to a version that resolves this vulnerability.

    Fixed in 2.3.4
  4. Upgrade

    Upgrade pip/tensorflow-cpu to a version that resolves this vulnerability.

    Fixed in 2.5.1
  5. Upgrade

    Upgrade pip/tensorflow-cpu to a version that resolves this vulnerability.

    Fixed in 2.4.3
  6. Upgrade

    Upgrade pip/tensorflow-cpu to a version that resolves this vulnerability.

    Fixed in 2.3.4
  7. Upgrade

    Upgrade pip/tensorflow to a version that resolves this vulnerability.

    Fixed in 2.5.1
  8. Upgrade

    Upgrade pip/tensorflow to a version that resolves this vulnerability.

    Fixed in 2.4.3
  9. Upgrade

    Upgrade pip/tensorflow to a version that resolves this vulnerability.

    Fixed in 2.3.4
  10. Upgrade

    Upgrade TensorFlow to a version that resolves this vulnerability.

    Fixed in 2.6.0Patch 23d6383eb6c14084a8fc3bdf164043b974818012
  11. Upgrade

    Upgrade TensorFlow to a version that resolves this vulnerability.

    Fixed in 2.5.1Patch 23d6383eb6c14084a8fc3bdf164043b974818012
  12. Upgrade

    Upgrade TensorFlow to a version that resolves this vulnerability.

    Fixed in 2.4.3
  13. Upgrade

    Upgrade TensorFlow to a version that resolves this vulnerability.

    Fixed in 2.3.4
  14. Configuration

    Do not use YAML deserialization paths that rely on `yaml.unsafe_load` for Keras model loading; ensure YAML model deserialization is performed using the patched/safe implementation.

    TensorFlow/Keras model YAML loading yaml.unsafe_load = disable or avoid use of yaml.unsafe_load when deserializing Keras models from YAML
  15. Compensating control

    Given that YAML format support requires a significant amount of work, the text indicates YAML format support has been removed for now—remove/disable YAML support for Keras model deserialization in your environment until patched.

Event History

Aug 12, 2021
CVE Published
via MITRE·11:05 PM
Data Sourced
via MITRE·11:05 PM
DescriptionSeverityWeakness
Data Sourced
via NVD·11:15 PM
RemedyDescriptionSeverityWeaknessAffected Software
Aug 25, 2021
Advisory Published
via GitHub·02:41 PM
Free Weekly Intel

Don't miss critical vulnerabilities

Join thousands of security professionals who receive our weekly digest of trending CVEs, zero-days, and exploited vulnerabilities.

No spam. Unsubscribe anytime.

Frequently Asked Questions

1

What is the severity of CVE-2021-37678?

CVE-2021-37678 is classified as a high-severity vulnerability due to its potential for arbitrary code execution.

2

How do I fix CVE-2021-37678?

To remediate CVE-2021-37678, upgrade TensorFlow to versions 2.5.1 or later.

3

Which versions of TensorFlow are affected by CVE-2021-37678?

CVE-2021-37678 affects TensorFlow versions prior to 2.5.1, including 2.5.0, 2.4.x, and 2.3.x.

4

What type of vulnerability is CVE-2021-37678?

CVE-2021-37678 is a deserialization vulnerability that could allow arbitrary code execution.

5

Can Keras models trigger CVE-2021-37678?

Yes, CVE-2021-37678 can be triggered when deserializing Keras models from YAML format.

Contact

SecAlerts Pty Ltd.
132 Wickham Terrace
Fortitude Valley,
QLD 4006, Australia
info@secalerts.co
By using SecAlerts services, you agree to our services end-user license agreement. This website is safeguarded by reCAPTCHA and governed by the Google Privacy Policy and Terms of Service. All names, logos, and brands of products are owned by their respective owners, and any usage of these names, logos, and brands for identification purposes only does not imply endorsement. If you possess any content that requires removal, please get in touch with us.
© 2026 SecAlerts Pty Ltd.
ABN: 70 645 966 203, ACN: 645 966 203