Open Internet by MindsNet

Vulnerabilities in AI Infrastructure

The research identified several vulnerabilities in Ollama and LiteLLM, including denial of service, resource exhaustion, authentication bypass, SSRF, and sandbox escape scenarios. These issues originated from often-overlooked areas in AI infrastructure, such as model parsing and conversion pipelines, resource allocation controls, authentication logic, network trust boundaries, and Unicode normalization edge cases.

Computing & Technology, Computer Science, Programming Languages