Unsloth re-checks models before every run

🔥 What happened Unsloth published a security overview for its Studio desktop app that re-checks models before every run. The key twist: a saved approval no longer counts if the repository code changes – the app re-scans and asks for fresh consent. 💡 Why it matters Two incidents drove this: compromised LiteLLM versions on PyPI and an infostealer hidden in a Hugging Face repo that hit #1 trending with a claimed 244,000 downloads. Unsloth now runs four gates: code fingerprinting, a separate weight-file check, OS sandboxing, and package scanning. Even top models like DeepSeek-OCR and Kimi-VL get flagged. ⚡ Our take A long-overdue reality check for the 'trusted publisher' fantasy. If you're pulling models from random repos, you shouldn't feel safe for a second – Unsloth just made that mandatory.