Advisory

PYSEC-2026-1043

Also known as: BIT-tensorflow-2022-35987, CVE-2022-35987, GHSA-w62h-8xjm-fv49, PYSEC-2026-3252, PYSEC-2026-3377

Summary

TensorFlow vulnerable to CHECK fail in DenseBincount

Details

Impact
DenseBincount assumes its input tensor weights to either have the same shape as its input tensor input or to be length-0. A different weights shape will trigger a CHECK fail that can be used to trigger a denial of service attack.

import tensorflow as tf
binary_output = True
input = tf.random.uniform(shape=[0, 0], minval=-10000, maxval=10000, dtype=tf.int32, seed=-2460)
size = tf.random.uniform(shape=[], minval=-10000, maxval=10000, dtype=tf.int32, seed=-10000)
weights = tf.random.uniform(shape=[], minval=-10000, maxval=10000, dtype=tf.float32, seed=-10000)
tf.raw_ops.DenseBincount(input=input, size=size, weights=weights, binary_output=binary_output)

Patches
We have patched the issue in GitHub commit bf4c14353c2328636a18bfad1e151052c81d5f43.

The fix will be included in TensorFlow 2.10.0. We will also cherrypick this commit on TensorFlow 2.9.1, TensorFlow 2.8.1, and TensorFlow 2.7.2, 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 Di Jin, Secure Systems Labs, Brown University

Severity

Medium Basis: CVSS v3 score 5.9, vector CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:N/A:H

Dates

Published
2026-07-07 10:17 UTC
Modified
2026-07-13 16:43 UTC

Affected packages and fixes

tensorflow-gpu

PyPI · All systems

Affected

  • introduced 0, fixed 2.7.2, introduced 2.8.0, fixed 2.8.1, introduced 2.9.0, fixed 2.9.1 (ecosystem)

Fixed versions

2.7.2, 2.8.1, 2.9.1

Fix

  • Fixed version 2.7.2
    pip install --upgrade tensorflow-gpu==2.7.2
  • Fixed version 2.8.1
    pip install --upgrade tensorflow-gpu==2.8.1
  • Fixed version 2.9.1
    pip install --upgrade tensorflow-gpu==2.9.1

Fix commands are the export's own generic commands. Ojuju never runs them. Check them against your setup first.

Check your own computer: guard packages check

View this record on OSV.dev

Source

Source: PyPA Advisory Database · CC BY 4.0

Sources & licences