Kristen Smith 
- Federal leaders flagged AI-accelerated vulnerability exploitation as the top threat, with just 36 percent confident their defenses could keep pace
- 22 percent said their organizations have not settled who is accountable when an AI agent causes a security incident
- The survey, run with Market Connections in April, appears in Booz Allen’s Velocity publication
Federal agencies are moving agentic artificial intelligence into their operations well ahead of their confidence in their ability to secure it, according to a Booz Allen Hamilton survey released Tuesday. Of more than 100 federal IT and cybersecurity decision-makers polled, 58 percent said their agencies have deployed or are piloting AI agents, but only 28 percent expressed high confidence in their ability to field those systems securely.
The survey appears in the latest issue of Velocity, Booz Allen’s emerging technology publication.
What Security Gaps Did the Booz Allen AI Survey Identify?
Concerns about AI-driven threats ran high, with 79 percent of respondents very or extremely worried that adversaries would use AI to accelerate cyberattacks over the next 12 to 18 months. Federal leaders pointed to AI-accelerated vulnerability exploitation — the shrinking window between discovering and exploiting a software flaw — as the top threat. Only 36 percent were confident their defenses could keep pace with AI-enabled attackers, and 31 percent felt fully or substantially prepared to run AI-powered defenses alongside their existing security tools.
Who Is Accountable When an AI Agent Fails?
Accountability emerged as an open question: 22 percent of respondents said their organizations have not settled who is responsible when an AI agent causes a security incident or operational failure. On deployment concerns specifically, respondents most often cited protecting sensitive or classified data, followed by preventing unauthorized actions, and guarding against adversarial manipulation and prompt injection. Leaders said clearer visibility into agent behavior, proven risk frameworks and demonstrated success in their own environments would make them more comfortable expanding agentic AI use.
Booz Allen and Market Connections conducted the online survey of 105 federal IT and cybersecurity decision-makers in April.
