Federal agencies are rapidly embracing agentic AI to automate decision-making and strengthen mission operations. However, a new Booz Allen survey reveals that while adoption is accelerating across government organizations, confidence in securely deploying these autonomous AI systems remains significantly lower.
Published in the latest edition of Velocity, Booz Allen's technology and mission innovation publication, the report highlights the growing need for robust AI governance, cybersecurity frameworks, identity management, continuous monitoring, red teaming, formal validation, and zero trust principles throughout the AI lifecycle. The findings emphasize that secure AI deployment will be essential as government agencies increasingly rely on agentic AI for mission-critical operations.
The survey, conducted among more than 100 federal IT and cybersecurity decision-makers, found that 58% of respondents have already deployed or are actively piloting agentic AI systems. Despite this momentum, only 28% expressed high confidence in their agencies' ability to implement these systems securely.
The findings illustrate the growing challenge for federal cybersecurity teams as they seek to leverage autonomous AI capabilities while maintaining security, governance, and operational accountability.
Cybersecurity concerns remain one of the largest barriers to wider agentic AI adoption. According to the survey, 79% of respondents reported being "very" or "extremely concerned" that adversaries will use AI to accelerate cyberattacks against federal agencies during the next 12 to 18 months.
Federal leaders identified AI-accelerated vulnerability exploitation as the most significant emerging threat, where attackers can dramatically reduce the time between identifying software vulnerabilities and exploiting them.
Confidence in current cyber defenses also remains limited. Only 36% of respondents believe existing cybersecurity capabilities can keep pace with increasingly sophisticated AI-enabled attackers.
While organizations recognize the value of AI-driven cybersecurity, implementation readiness remains low. Only 31% of surveyed federal cyber and IT leaders believe they are fully or substantially prepared to deploy AI-powered cyber defenses that integrate with existing security infrastructure.
The report suggests that successful AI adoption requires more than deploying intelligent agents. Organizations must also establish comprehensive governance models, identity controls, continuous monitoring, and formal validation processes to manage autonomous AI safely.
Survey respondents identified several major concerns surrounding agentic AI deployments. Protecting sensitive or classified information ranked highest, followed by preventing unauthorized AI actions and ensuring resilience against adversarial manipulation and prompt injection attacks.
The research also revealed ongoing uncertainty around accountability. Nearly one-quarter of respondents indicated their organizations have not clearly defined responsibility when an AI agent causes a cybersecurity incident or operational failure.
Federal leaders believe broader adoption of agentic AI will depend on stronger operational visibility and proven governance models. Respondents identified greater visibility into AI agent behavior, established risk mitigation frameworks, and demonstrated success within their own operating environments as the primary factors that would increase confidence in expanding AI deployments.
The findings reinforce that while agentic AI offers significant operational advantages, secure implementation, transparent governance, and cybersecurity resilience will determine how quickly federal agencies can scale these technologies across mission-critical environments.
Booz Allen is an advanced technology company. We build commercial-grade products and solutions for America’s most critical defense, civil, and national security priorities. For more information, visit www.boozallen.com.