Your RSA-2048 keys break in 2030. Find every one of them before attackers do.

CVE-2026-5241

CRITICAL

A vulnerability in the LightGlue model loading path of huggingface/transformers version 5.2.0 allows an attacker-controlled model repository to execute arbitrary code during model initialization.…

Published
Jun 3, 2026
Updated
Jun 17, 2026
Affected
0 pkgs
Patched
None yet
Exploits
None indexed

EPSS Exploitation Probability

via FIRST.org ↗
0.4%probability of exploitation in next 30 days
Lower Risk27th percentile0.00%

EPSS (Exploit Prediction Scoring System) is a daily probability model maintained by FIRST.org. It estimates the likelihood a CVE will be exploited in production environments within the next 30 days, derived from real-world threat intelligence signals.

Description

A vulnerability in the LightGlue model loading path of huggingface/transformers version 5.2.0 allows an attacker-controlled model repository to execute arbitrary code during model initialization. The issue arises because the trust_remote_code parameter, intended to prevent remote code execution, is overridden by untrusted serialized configuration data in a nested code path. Specifically, when loading a LightGlue model using AutoModel.from_pretrained() with trust_remote_code=False, the LightGlueConfig reads the trust_remote_code value from the untrusted config.json file and propagates it into nested AutoConfig.from_pretrained() calls. This results in the execution of attacker-provided Python modules, even when the victim explicitly disables remote code execution. The vulnerability poses a high risk for environments such as API inference servers, research notebooks, CI/CD pipelines, and model evaluation workers, potentially leading to credential theft, lateral movement, or persistence/backdoor deployment.

Affected Products

1 product · 1 configurations
Application
transformershuggingface
1 version
5.2.0

Detection & mitigation playbook

Vendor / appliance
  1. Detect

    Inventory every huggingface transformers deployment and check each version against the affected-products list above. Because the exploit targets the running system rather than your application code, also watch for remote code execution at the network and runtime layer — O3 flags the exploit behaviour from runtime telemetry and egress traffic even before a vulnerable build is confirmed.

  2. Remediation status

    No patch has shipped for CVE-2026-5241 yet — track the huggingface transformers advisory for a fixed release and apply the workarounds below in the meantime.

  3. Mitigate without a patch

    Cut exposure now: restrict the management/admin interface to trusted networks, segment the device, and apply the vendor's recommended configuration mitigations and any WAF/IPS signature. O3's runtime protection blocks the exploit chain at execution, holding the line on unpatched or end-of-life systems until you can patch.

  4. How O3 protects you

    O3 detects and blocks CVE-2026-5241 exploitation at runtime: eBPF exploit-chain detection, plus L7 egress monitoring that catches the post-exploitation callback and severs the attacker's outbound channel.

Tailored to CVE-2026-5241. Runtime protection reduces exposure until a permanent patch is applied and verified — it complements patching, it doesn't replace it.

Frequently Asked Questions

A vulnerability in the LightGlue model loading path of huggingface/transformers version 5.2.0 allows an attacker-controlled model repository to execute arbitrary code during model initialization. The issue arises because the `trust_remote_code` parameter, intended to prevent remote code execution, is overridden by untrusted serialized configuration data in a nested code path. Specifically, when loading a LightGlue model using `AutoModel.from_pretrained()` with `trust_remote_code=False`, the `LightGlueConfig` reads the `trust_remote_code` value from the untrusted `config.json` file and propagat
O3 Security · Runtime Protection

Is CVE-2026-5241 being exploited in your environment?

O3's eBPF runtime sensors and L7 egress monitoring detect and block the CVE-2026-5241 exploit chain at execution — protecting unpatched and end-of-life systems until the vendor patch is applied.