Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

NewsResearch

Unsloth Studio ran code from a model's config just by inspecting it

Pillar Security found that Unsloth Studio's web UI loaded model configurations with trust_remote_code enabled, so merely inspecting a malicious model on Hugging Face ran the Python code it shipped, with no weights downloaded and no inference. It was reported in June and fixed in Unsloth 2026.6.9; details were published on September 29.

Why it matters

Looking at a model can be enough to run its code. Keep trust_remote_code off everywhere, including tools that only "preview" models.

Sources

Read it with the rest of the news in the Learning Center