PT-2026-59763

Published

2026-07-09

·

Updated

2026-07-09

Description by

CVSS 3.1
8.8
High
MetricsAV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H

Impact

An attacker can craft a TFLite model that would cause an integer overflow [in embedding lookup operations](https://github.com/tensorflow/tensorflow/blob/ca6f96b62ad84207fbec580404eaa7dd7403a550/tensorflow/lite/kernels/embedding lookup sparse.cc#L179-L189):
cc
 int embedding size = 1;
 int lookup size = 1;
 for (int i = 0; i < lookup rank - 1; i++, k++) {
  const int dim = dense shape->data.i32[i];
  lookup size *= dim;
  output shape->data[k] = dim;
 }
 for (int i = 1; i < embedding rank; i++, k++) {
  const int dim = SizeOfDimension(value, i);
  embedding size *= dim;
  output shape->data[k] = dim;
 } 
Both embedding size and lookup size are products of values provided by the user. Hence, a malicious user could trigger overflows in the multiplication.
In certain scenarios, this can then result in heap OOB read/write.

Patches

The fix will be included in TensorFlow 2.8.0. We will also cherrypick this commit on TensorFlow 2.7.1, TensorFlow 2.6.3, and TensorFlow 2.5.3, 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 Wang Xuan of Qihoo 360 AIVul Team.
Exploitability
Fix
Vendors
Affected products
Related identifiers
PYSEC-2026-3146
Found an issue in the description? Have something to add? Feel free to write us 👾