Local process abort via live allocation deletion
Published Apr 3, 2025 · Updated Apr 3, 2025
Memory corruption in PyTorch 2.6.0 allows local users to interrupt a process by deleting a CUDA allocation that remains in use. torch.cuda.memory.caching_allocator_delete passes a live tensor pointer to the CUDA caching allocator's deletion path while the tensor still owns that allocation, causing an abort. Reachability requires local Python execution and an NVIDIA CUDA device; the published demonstration terminates only the invoking process, and PyTorch classifies invalid-argument crashes as ordinary bugs rather than security vulnerabilities.
Summary
What happened
Memory corruption in PyTorch 2.6.0 allows local users to interrupt a process by deleting a CUDA allocation that remains in use. torch.cuda.memory.caching_allocator_delete passes a live tensor pointer to the CUDA caching allocator's deletion path while the tensor still owns that allocation, causing an abort. Reachability requires local Python execution and an NVIDIA CUDA device; the published demonstration terminates only the invoking process, and PyTorch classifies invalid-argument crashes as ordinary bugs rather than security vulnerabilities.
The record
- CVE
- CVE-2025-3136
- Published
- Apr 3, 2025
- Updated
- Apr 3, 2025
- Vendor
- PyTorch
- Product
- PyTorch
- Classifications
- CWE-119
- Attack vector
- local
- Privileges
- authenticated
Timeline
How it unfolded
- Apr 3, 2025CVE publishedPublication date reported by the CVE source.
- Apr 3, 2025Record updatedLatest update available in the CVE record.
Exploitability
Present is not the same as exploitable
Compare your product and version with the public record. A matching version still requires validation against your environment.
Is a vulnerable build present?
Compare these published version ranges with your installed build and any vendor patches.
- Affected versionversion=2.6.0
What conditions does exploitation require?
What is affected?
Published CVSS scores
CVSS describes severity. EPSS estimates exploitation probability.
Attacks
What attackers are doing with it
Daily unique IPs observed by Shadowserver honeypots for known exploited vulnerabilities (KEVs). Missing observations do not establish an absence of attacks.
Weakness, pattern, technique
Public exploit references
- CUDA allocation abort reproducerproof of concept · demonstrated
Labels summarize the accepted research assessment. They do not indicate a test against your environment.
Technologies
Your stack
See the directory against your own environment.
Your stack
Check the software in your environment
Book a demo to see how Hinoki identifies affected software and validates exploitability in your environment.
Book a demo