Federal AI Security: Agencies Lack Trust in Own Systems

Few federal agencies trust the security of their own AI systems. That's according to a new survey by Booz Allen Hamilton. Here's the thing: many agencies are testing autonomous artificial intelligence, but less than one-third of federal tech leaders are confident in deploying them securely. While 51% are piloting these systems, only 7% have them actively in production. Another 22% plan to deploy them within the next year. Only 28% of those surveyed are "extremely" or "very" confident in their agency's ability to deploy agentic AI securely. Half are moderately confident, and 23% have little to no confidence. What's interesting is that AI agents can plan tasks and take autonomous actions across systems with limited human involvement, unlike generative AI chatbots. This poses security risks because agents could use valid permissions in unintended ways. Protecting sensitive data is the top concern, cited by 56% of respondents. Other worries include unauthorized agent actions and adversarial manipulation. The bottom line is that applying traditional security controls to AI agents that make their own decisions is challenging. This matters because the secure deployment of AI agents could significantly impact government operations and data.