-
CVSS Score
-The vulnerability stems from the SVD GPU kernel implementation. The commit diff shows a critical addition checking if shapeRaw.num_elements() == 0 to early-exit, addressing empty input handling. Since tf.linalg.matrix_rank relies on SVD internally, and the test case demonstrates triggering via matrix_rank, the vulnerable code path is in the SvdOpGpu's ComputeAsync method where the missing empty batch check caused the fatal assertion.
| Package Name | Ecosystem | Vulnerable Versions | First Patched Version |
|---|---|---|---|
| tensorflow | pip | < 2.7.2 | 2.7.2 |
| tensorflow | pip | >= 2.8.0, < 2.8.1 | 2.8.1 |
| tensorflow | pip | >= 2.9.0, < 2.9.1 | 2.9.1 |
| tensorflow-cpu | pip | < 2.7.2 | 2.7.2 |
| tensorflow-cpu | pip | >= 2.8.0, < 2.8.1 | 2.8.1 |
| tensorflow-cpu | pip | >= 2.9.0, < 2.9.1 | 2.9.1 |
| tensorflow-gpu | pip | < 2.7.2 | 2.7.2 |
| tensorflow-gpu | pip | >= 2.8.0, < 2.8.1 | 2.8.1 |
| tensorflow-gpu | pip | >= 2.9.0, < 2.9.1 | 2.9.1 |
A Semantic Attack on Google Gemini - Read the Latest Research