CVE-2026-33626: How attackers exploited LMDeploy LLM Inference Engines in 12 hours

Attackers weaponized a newly disclosed SSRF vulnerability in LMDeploy’s LLM inference engine within just 12 hours of advisory publication, using it to probe internal cloud metadata, databases, and admin interfaces—demonstrating how AI infrastructure bugs are now exploited almost immediately, often without any public exploit code.