HighVulnerability
CVE-2020-15193: Tensorflow dlpack.to_dlpack uses uninitialized memory when passed a non-tensor
- Source
- NVD(opens in a new tab)
- Identifier
- CVE-2020-15193
- Published
- Record updated
- Affected
- tensorflow = 2.2.0, fixed in 2.2.1
- tensorflow = 2.3.0, fixed in 2.3.1
- tensorflow-cpu = 2.2.0, fixed in 2.2.1
- and 3 more
- Known exploitation
- Not listed in the CISA Known Exploited Vulnerabilities catalog at the last check.
- EPSS
- 0.8%
Summary
TensorFlow versions before 2.2.1 and 2.3.1 have a flaw in `dlpack.to_dlpack`. The pybind11 glue code assumes its argument is a tensor, but nothing stops callers from passing a Python object, so a `reinterpret_cast` produces a pointer to uninitialized memory, leading to further memory corruption.
Mitigation
The issue is patched in commit 22e07fb204386768e5bcbea563641ea11f96ceb8 and released in TensorFlow versions 2.2.1 and 2.3.1.
Related items
- CriticalCVE-2026-90970: GitLab AI Gateway prompt template sandbox escape via crafted flow configurationSame vendor · NVD/CVE Database
- MediumCVE-2026-89278: GPTranslate WordPress plugin exposes plaintext API key to unauthenticated usersSame vendor · NVD/CVE Database
- CriticalCVE-2026-19407: Google Cloud Gemini Enterprise Agent Platform SDK for Python bucket squattingSame vendor · NVD/CVE Database
- HighCVE-2026-19486: A Server-Side Request Forgery (SSRF) vulnerability in Google Cloud Gemini Enterprise Agent Platform App Builder…Same vendor · NVD/CVE Database
- HighCVE-2026-13745: Gemini CLI code execution via untrusted .env files overriding GEMINI_CLI_HOMESame vendor · NVD/CVE Database