Security readout for executives and security teams
Plain-English summary
A malicious Keras model saved in YAML format could make affected TensorFlow/Keras versions run arbitrary code during loading. The business risk is highest where ML teams import models, templates, or artifacts from third parties, shared storage, or automated pipelines without trust controls.
Executive priority
Treat this as high-priority for ML platforms handling shared or third-party model artifacts. It enables code execution with high impact, but urgency should be highest where YAML Keras models are loaded from sources that are not fully trusted.
Technical view
CVE-2021-37678 is a CWE-502 unsafe deserialization issue. TensorFlow/Keras used yaml.unsafe_load while deserializing Keras YAML model configuration, allowing arbitrary code execution from crafted input. The vendor removed YAML format support and patched via commit 23d6383eb6c14084a8fc3bdf164043b974818012, with fixed releases planned for 2.6.0 and supported backports.
Likely exposure
Exposure is likely in Python ML environments using TensorFlow/Keras versions before 2.6.0, 2.5.1, 2.4.3, or 2.3.4, especially where YAML model files are loaded. Internet-facing exposure is not established by the sources; the CVSS vector lists local attack vector.
Exploitation context
The provided sources do not show active exploitation, and the CVE is not marked KEV. Exploitation requires an affected environment to deserialize a crafted Keras model from YAML. The practical risk depends on model artifact provenance and whether untrusted YAML models are accepted.
Researcher notes
The key evidence is the vendor advisory: yaml.unsafe_load was used in Keras model_config YAML deserialization, and YAML support was removed as the fix. Validate affected branches against the stated version ranges and commit rather than assuming all TensorFlow deployments are reachable.
Mitigation direction
- Upgrade TensorFlow to 2.6.0 or the supported fixed backport for your branch.
- Stop loading Keras models from YAML in affected environments.
- Restrict model artifact sources to trusted repositories and controlled pipelines.
- Check TensorFlow vendor guidance before relying on alternate mitigations.
Validation and detection
- Inventory TensorFlow and Keras versions across notebooks, training jobs, APIs, and batch workers.
- Identify code paths or workflows that deserialize Keras models from YAML.
- Review ML artifact stores for externally supplied or untrusted YAML model files.
- Confirm upgraded environments no longer rely on vulnerable YAML deserialization support.
Public sources used
Generated from the cited source records. This long-tail analysis has not been individually reviewed by a named human.
Conservative CVE-to-ATT&CK context
These mappings and lookup hints may be relevant to the vulnerability behavior, CWE, affected product, or exposure path. Glexia-inferred context is not an official MITRE, ATT&CK, CWE, or CVE Program mapping.
ATT&CK lookup starting points
Use these exact CWE pages and searches to review the Glexia ATT&CK library from this CVE's weakness and description context.
CWE-502: Code execution behavior lookup
Code execution and unsafe deserialization weaknesses often justify reviewing execution behavior and process telemetry. Open the exact CWE lookup page first, then review the ATT&CK searches from that MITRE weakness context. This is a Glexia lookup hint, not an official ATT&CK mapping.
Open ATT&CK lookupExecution behavior lookup
The CVE wording references code or command execution, so execution technique review may help defensive triage. This is a Glexia inferred lookup path, not an official MITRE, ATT&CK, or CVE Program mapping.
Open ATT&CK lookupCVE-2021-37678 mapping review
Open the CVE-to-ATT&CK bridge for reviewed, inferred, or future official mappings tied to this CVE.
Open ATT&CK lookup- Severity
- Critical
- CVSS
- 9.3 (3.1)
- Known Exploited
- No
- Published
Vector: CVSS:3.1/AV:L/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H
CNA and ADP enrichment extracted from CVE v5
These fields come from the CVE record and ADP containers, not from Glexia's Take. They preserve time-varying source decisions such as CISA SSVC, KEV status, CVSS metrics, and provider references.
CVSS vector scores
1 official scoreWe collect every scored CVSS vector available in the official CNA and ADP containers. When more than one version is present, the table keeps the source vectors side by side instead of collapsing them into the highest score.
CVSS:3.1/AV:L/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H2.56Primary CVE scoreVulnerability scoring details
Base CVSS 3.1 score
9.3CriticalVector: CVSS:3.1/AV:L/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H
Source materials
- CVE List V5 sourceCVE List V5
- https://github.com/tensorflow/tensorflow/security/advisories/GHSA-r6jx-9g48-2r5rCVE reference · x_refsource_CONFIRM
- https://github.com/tensorflow/tensorflow/commit/23d6383eb6c14084a8fc3bdf164043b974818012CVE reference · x_refsource_MISC
Products and packages named in the record
CWE details
CWE links open Glexia weakness intelligence pages with official CWE context, developer remediation guidance, and related CVE mappings.
Deserialization of Untrusted Data
Deserialization of Untrusted Data represents a recurring weakness pattern that can create exploitable paths when design, validation, or implementation controls are missing.