Hackers Launch Over 91,000 Attacks on AI Systems Using Fake Ollama Servers

GreyNoise researchers observed two active campaigns targeting AI infrastructure between October 2025 and January 2026: one using Server-Side Request Forgery (SSRF) to make AI servers call attacker-controlled hosts (with a spike of 1,688 sessions over 48 hours), and a second large-scale reconnaissance campaign that generated ~80,469 sessions from two IPs probing over 73 AI endpoints (including OpenAI-compatible, Google Gemini formats, Anthropic, Meta Llama, xAI Grok, and others) to build target lists; the report lists source IPs, session counts, and recommends restricting model downloads to trusted sources and monitoring rapid, repetitive queries.