Cyber Risks · The Hacker News

We Scanned 1 Million Exposed AI Services. Here's How Bad the Security Actually Is

Summary: Intruder used certificate transparency logs to scan just over 2 million hosts and 1 million exposed AI services, finding the AI infrastructure more vulnerable, exposed, and misconfigured than any other software the firm has investigated. Many hosts ran out of the box with no authentication because authentication is not enabled by default in many of these projects, and of 5,200+ exposed Ollama API servers queried with a single prompt, 31% answered without requiring authentication.

Key Takeaway: Over 90 exposed chatbot instances spanned government, marketing, and finance, with workflows, prompts, and outward access open to anyone - and 518 of the models found were wrapping paid frontier models from Anthropic, Deepseek, Moonshot, Google, and OpenAI, meaning attackers can use someone else's infrastructure and credentials to reach capable models without paying or being logged.

Read the original at The Hacker News ↗