Patched Unsloth Studio flaw could run code during model inspection

Pillar Security said selecting a model could expose data and credentials accessible to the application. The identified fix was released on June 18, 2026.

Pillar Security said selecting a model could expose data and credentials accessible to the application. The identified fix was released on June 18, 2026.
Selecting a model in Unsloth Studio could cause the application to download and run Python code from that model’s Hugging Face repository, according to Pillar Security’s finding described by The Hacker News on October 1, 2026. The code ran during inspection, before the application loaded the model or used it to produce outputs.
The Hacker News reported that the issue was addressed in version 2026.6.9, released on June 18, 2026.
Unsloth is an open-source library for fine-tuning large language models—adapting them through additional training—and quantizing them, which reduces the numerical precision used to represent them.
Pillar Security said Studio’s model picker executed code while checking metadata, the information describing a model. The application did not need to load the model’s weights—the numerical values learned during training—or run inference, meaning use the model to produce an output. Selecting the model was enough to trigger execution of code stored alongside its files.
Pillar Security said an attacker could run code as the user, potentially stealing accessible data, altering models and training outputs, or using available credentials to reach other systems. Potentially exposed assets included proprietary training data, saved model files, Hugging Face access tokens, SSH keys used to authenticate connections, and cloud credentials. That access was limited to what the running process could reach.
It is not yet clear which versions were affected, when Pillar Security disclosed the flaw or whether anyone exploited it.

The findings describe access information left public for years, but do not establish whether anyone misused it.
2 min

The Guardian reported a threat to Oxbridge technology-hub funding and a warning about UK–UAE relations following Manchester City’s financial-rule verdict.

Automakers are weighing central processors against specialized chips distributed through a car. Moving sensor data and checking software updates over years of use are equally important challenges.