Trusting Unsloth’s trust setting could mean executing malicious code

A high-severity vulnerability in Unsloth’s model loader could allow arbitrary Python code execution when processing a maliciously crafted model configuration, CSO Online reports. Tracked as CVE-2026-93348 and rated CVSS 8.6, the flaw disclosed by Pillar Security means AI developers may have a security setting enabled and still end up running attacker-controlled code.