Censys has found a 63% increase in exposed AI infrastructure over a nine-month period. The security firm's data points to a rapid expansion of AI systems that are left open to the internet without proper safeguards. That combination—fast deployment and weak security—is making it easier for attackers to find and target vulnerable components.
The nine-month surge in exposed services
The 63% jump didn't happen overnight. Over nine months, the number of AI-related services and endpoints visible to anyone with an internet connection climbed steadily. Censys tracks internet-wide scan data, and its researchers flagged the rise as part of their ongoing monitoring of exposed assets. The company didn't name specific victims or industries, but the trend spans cloud-hosted tools, model-serving frameworks, and development environments that were never meant to be public.
Why rapid AI rollouts leave gaps
Companies are under pressure to ship AI features, and security often gets treated as a later step. When teams spin up GPU clusters, inference APIs, or vector databases, they don't always lock down access controls or authentication. The result is a growing attack surface: exposed dashboards, unauthenticated endpoints, and misconfigured storage buckets. Censys's finding suggests that as AI adoption accelerates, the number of these weak points is growing just as fast. The firm stressed that deploying AI without adequate security measures directly increases vulnerability.
The urgent need for cybersecurity focus
Censys called for an urgent shift toward cybersecurity to reduce the risks from this exposed infrastructure. That means treating AI systems like any other critical asset—with regular scanning, strict access policies, and prompt patching. The company's message is blunt: the window to secure these systems before they're exploited is narrowing. With 63% more exposed services than nine months ago, the problem is already large enough to warrant immediate attention.
For security teams, the next step is clear: find out how much of your AI infrastructure is visible from the outside. Censys and similar tools can help map that exposure, but the fix—closing unnecessary ports, adding authentication, and segmenting networks—still falls to the organizations running the systems. The 63% figure is a warning, not a prediction, and it's up to defenders to act on it before attackers do.
