United States
Australia
United Kingdom
Europe
New Zealand
Canada
Federal agencies are accelerating the use of autonomous AI agents in mission environments even as confidence in securing those systems lags, a new Booz Allen survey shows.
In findings released July 21 from more than 100 federal IT and cybersecurity decision-makers, 58% said their agencies have deployed or are piloting AI agents, while only 28% expressed high confidence in their ability to deploy them securely. Concern about AI-enabled cyber threats remains elevated: 79% reported being “very” or “extremely concerned” about adversaries using AI to speed cyberattacks over the next 12 to 18 months.
Respondents identified AI-accelerated vulnerability Exploitation as the most significant AI-enabled cyber threat, warning that machine-driven tools can dramatically compress the window between discovering and exploiting software flaws. Just 36% said they are confident current defenses can keep pace with AI-enabled attackers, and only 31% reported being fully or substantially prepared to field AI-powered cyber defenses that integrate with existing security infrastructure.
Agencies flagged several risks tied to deploying agentic AI, including protecting sensitive or classified data (56%), preventing unauthorized actions (50%), and ensuring resilience against adversarial manipulation and prompt injection attacks (37%). Accountability remains unsettled: 22% said their organizations have not clearly determined who is responsible if an AI agent triggers a security incident or operational failure.
Federal leaders pointed to steps that would bolster confidence in expanding agentic AI: greater visibility into agent behavior (56%), proven risk mitigation frameworks (44%), and demonstrated success within their own environments (42%).
The results appear in the latest issue of Velocity, Booz Allen’s publication on emerging technology and mission innovation. This edition details a systems-level approach to deploying agentic AI securely, including identity governance, continuous monitoring, red teaming, formal validation, and zero trust principles across the AI lifecycle.
The survey was conducted online in April 2026 by Market Connections and Booz Allen, polling 105 federal government decision makers and influencers involved in IT and cybersecurity strategy.
