Why it's so easy to find open, exposed AI servers on the web

Cisco security researchers discovered over 1,100 exposed Ollama AI servers online, with 20% hosting vulnerable models. These unsecured servers risk high costs, hardware damage, model theft, and misuse. AI adoption has outpaced security practices, often prioritizing accessibility over protection. Tailscale CEO Avery Pennarun advises against public-facing private API servers, noting AI companies often neglect security. Solutions like Tailscale can host AI servers securely, invisible to the public web but accessible to authorized users.