CVE-2026-24747Disclosure(linuxfoundation / pytorch)

MEDIUMCVSS 8.8 · HIGH

Exploit discussion active in current signal (2 latest mentions)

Immediate actions

Patch linuxfoundation pytorch systems immediately
Hunt for exploitation attempts and persistence artifacts
Increase monitoring for publicly documented tradecraft

Recommended action window: High priority (within 72h)

NVD description

PyTorch is a Python package that provides tensor computation. Prior to version 2.10.0, a vulnerability in PyTorch's `weights_only` unpickler allows an attacker to craft a malicious checkpoint file (`.pth`) that, when loaded with `torch.load(..., weights_only=True)`, can corrupt memory and potentially lead to arbitrary code execution. Version 2.10.0 fixes the issue.

linuxfoundation CVEs Back to dashboard

4.3/ 10 priority

Sources & remediation

Exploit / PoC references

https://github.com/pytorch/pytorch/issues/163105

Vendor / third-party advisories

https://github.com/pytorch/pytorch/security/advisories/GHSA-63cw-57p8-fm3p

Other references

Weakness type (CWE)

CWE-94CWE-502

Priority

MEDIUM

Exploitation

NONE

PoC

YES

Patch

AVAILABLE

Momentum

STABLE

Are you affected?

If you run products in this scope, you should treat this CVE as relevant to your environment.

pytorch

Threat summary

Public PoC and exploit tooling are both present
Patch or workaround signal is available
12 mentions across 8 observed days
Momentum state: stable

What's happening

Exploit tool or code specified in 1 signal
PoC mentioned or linked in 2 signals
Patch or workaround mentioned in 2 signals
Technical details provided in 10 signals
Disclosure: 5 classified signals
General: 3 classified signals
Peaked 6d ago at 3 mentions (2026-01-28); latest day: 2
12 total mentions across 8 days

Affected systems

Vendors

linuxfoundation

Products

pytorch

Deep dive

Activity timeline12 mentions / 8d

Signal classification5 categories

Disclosure

541.7%

General

325.0%

Patch

216.7%

PoC

18.3%

Exploit

18.3%

Referenced assets7 URLs

By indicator

Classification over time

Date	Total	Labels

2026-01-27	2	Disclosure2
2026-01-28	3	Disclosure1Patch2
2026-01-30	1	PoC1
2026-03-03	1	Disclosure1
2026-03-04	1	Exploit1
2026-03-07	1	General1
2026-07-01	1	Disclosure1
2026-09-01	2	General2

Full discourse12 posts

Giuseppe `N3mes1s`@N3mes1s215d
General
I found the same pattern for the CVE-2026-24747 RCE. I had to push opus 4.6 around 10 research sessions and multiple double check and pus for primitives and deep code understanding before be able to produce the exploit. I would say in a complex scenario is 10X complicated
Post summary
The author indicates their research led to an exploit for CVE-2026-24747, clarifying the issue as an RCE, but no PoC, code, or patch details are provided.
180834511.9K
12.8K followersView on X
Azrael@azraelxuemo251d
PoC
RCE video demo of cve-2026-24747 https://t.co/z2zKkqfkU6
Post summary
A video demo illustrating remote code execution for CVE‑2026‑24747 is shared, serving as a proof of concept without mentioning exploitation tools, patches, or active attacks.
08046203.7K
149 followersView on X
Giuseppe `N3mes1s`@N3mes1s218d
Exploit
And we have a fully weaponized CVE-2026-24747 The victim does only: torch.load(open("evil.pth","rb"), weights_only=True) #pruva https://t.co/r4piOmrYYB
Post summary
The tweet claims that CVE-2026-24747 is fully weaponized, providing a basic code snippet that demonstrates how to exploit the flaw via PyTorch's torch.load, but it offers no patch info or evidence of wild exploitation.
03034134.0K
12.8K followersView on X
Simone Margaritelli@evilsocket219d
Disclosure
AI ( Pruva ) is very close to fully weaponize CVE-2026-24747, with RCE on any service that supports uploading and running a PyTorch model, regardless of security hardening ( weights_only=True ).
Post summary
The entry announces that CVE-2026-24747 can lead to remote code execution on any service running PyTorch models, overriding security hardening settings.
01022113.6K
47.5K followersView on X
Azrael@azraelxuemo38d
General
The story behind the CVE-2026-24747 PyTorch weights_only bypass
Post summary
The snippet references CVE-2026-24747 involving a PyTorch weights_only bypass but provides no further details on exploitation, mitigation, or technical specifics.
6002162.1K
207 followersView on X
deepsec.cc@deepsec_cc99d
Disclosure
#deepsec "One Chain to Own Them All: Breaking AI Infrastructures". weights_only is the protection everyone leans on. @azraelxuemo & @llfamsec got past it — one heap overflow (CVE-2026-24747) hitting vLLM, ComfyUI, Elasticsearch & more, on default configs. https://deepsec.cc/202607/speaker-jian-zhou-lei-lu.html
Post summary
DeepSec disclosed a heap overflow (CVE-2026-24747) impacting multiple AI infrastructure tools on default configurations; no PoC, exploit code, or active exploitation details were provided.
0001141.3K
556 followersView on X
Azrael@azraelxuemo38d
General
1. When I discovered CVE-2026-24747 last year, I initially thought it wasn't very exploitable. Although it had a write primitive, it lacked a leak, so for a while I considered it a useless vulnerability.
Post summary
The author shares a personal assessment that CVE-2026-24747, while having a write primitive, may be non-exploitable due to lacking a memory leak, but no further details or exploit code are given.
00010192
203 followersView on X
PulsePatch.io@pulsepatchio253d
Patch
PyTorch is vulnerable to remote code execution via untrusted checkpoint files (CVE-2026-24747). Review your model loading practices and consider updating. #PyTorch #RCE #infosec https://www.pulsepatch.io/posts/cve-2026-24747-pytorch-remote-code-execution
Post summary
The post highlights a remote code execution flaw in PyTorch and advises users to update their installations to mitigate the risk.
0000095
1 followersView on X
PulsePatch.io@pulsepatchio253d
Patch
PyTorch versions prior to 2.10.0 are vulnerable to arbitrary code execution via malicious .pth checkpoint files (CVE-2026-24747). Load trusted models only. #PyTorch #MachineLearning #infosec https://www.pulsepatch.io/posts/pytorch-weights-only-unpickler-rce-cve-2026-24747
Post summary
PyTorch versions below 2.10.0 are vulnerable to arbitrary code execution when loading malicious .pth checkpoint files; users are advised to only load trusted models, with details hosted on PulsePatch.
0000085
1 followersView on X
Vulmon Vulnerability Feed@VulmonFeeds254d
Disclosure
CVE-2026-24747 Memory Corruption and Code Execution Vulnerability in PyTorch Bef... https://vulmon.com/vulnerabilitydetails?qid=CVE-2026-24747 Customizable Vulnerability Alerts: https://alerts.vulmon.com/?utm_source=twitter&utm_medium=social&utm_campaign=2102281&utm_content=4
Post summary
The tweet announces CVE-2026-24747, a memory corruption and code execution flaw in PyTorch, and links to a details page but provides no PoC, exploit, or patch information.
0000059
4.0K followersView on X
CVE@CVEnew254d
Disclosure
CVE-2026-24747 PyTorch is a Python package that provides tensor computation. Prior to version 2.10.0, a vulnerability in PyTorch's `weights_only` unpickler allows an attacker to cra… https://www.cve.org/CVERecord?id=CVE-2026-24747
Post summary
The CVE-2026-24747 entry highlights a flaw in PyTorch's weights_only unpickler, enabling malicious payload crafting before v2.10.0, with no evidence of active exploitation, patches, or PoC references.
00000296
56.5K followersView on X
cvereports@_cvereports254d
Disclosure
CVE-2026-24747: Heavy Weights: Crushing PyTorch's 'Secure' Loader via Heap Corruption A critical heap corruption vulnerability in PyTorch's restricted unpickler allows attackers to bypass the `weights_only=True` security flag, turning safe model loadi... https://cvereports.com/reports/CVE-2026-24747
Post summary
A critical heap corruption vulnerability in PyTorch's restricted unpickler has been disclosed, enabling attackers to bypass the weights_only security flag used for safe model loading.
0000059
29 followersView on X

CPE platform detail1 entries

Part:

Vendor:

Product:

Version:

Target SW:

Target HW:

1 of 1 entries

Part	Vendor	Product	Version	Target SW	Target HW
App	linuxfoundation	pytorch	-	python	-

Explore more

Back to dashboard linuxfoundation CVEs pytorch vulnerabilities

What researchers are saying

CVE-2026-24747

Giuseppe `N3mes1s`[verified]@N3mes1sGeneral

The author indicates their research led to an exploit for CVE-2026-24747, clarifying the issue as an RCE, but no PoC, code, or patch details are provided.

Giuseppe `N3mes1s`[verified]@N3mes1sExploit

The tweet claims that CVE-2026-24747 is fully weaponized, providing a basic code snippet that demonstrates how to exploit the flaw via PyTorch's torch.load, but it offers no patch info or evidence of wild exploitation.

Simone Margaritelli[verified]@evilsocketDisclosure

The entry announces that CVE-2026-24747 can lead to remote code execution on any service running PyTorch models, overriding security hardening settings.

Azrael@azraelxuemoPoC

A video demo illustrating remote code execution for CVE‑2026‑24747 is shared, serving as a proof of concept without mentioning exploitation tools, patches, or active attacks.

Azrael@azraelxuemoGeneral

The snippet references CVE-2026-24747 involving a PyTorch weights_only bypass but provides no further details on exploitation, mitigation, or technical specifics.

deepsec.cc@deepsec_ccDisclosure

DeepSec disclosed a heap overflow (CVE-2026-24747) impacting multiple AI infrastructure tools on default configurations; no PoC, exploit code, or active exploitation details were provided.

Azrael@azraelxuemoGeneral

The author shares a personal assessment that CVE-2026-24747, while having a write primitive, may be non-exploitable due to lacking a memory leak, but no further details or exploit code are given.

PulsePatch.io@pulsepatchioPatch

The post highlights a remote code execution flaw in PyTorch and advises users to update their installations to mitigate the risk.