Definition
An attack that weaponizes exposed, unauthenticated AI model servers (such as open-source Ollama instances) as the autonomous reasoning engine for multi-stage cyberattacks. Attackers use the compromised model to plan, execute, and refine attack strategies without their own computational resources.
Why it matters
Hundreds of thousands of AI model servers are publicly accessible with no authentication, providing attackers with free, sophisticated compute for launching advanced campaigns. A single exposed model becomes a persistent threat multiplier across an organization's entire attack surface.