CRS Report on Agentic Artificial Intelligence and Cyberattacks
The Government Accountability Office report examined agentic artificial intelligence capabilities. Agentic AI involves systems that accomplish specific goals with limited supervision. The report noted that there is no known official government guidance or policies on agentic AI.
Understanding Agentic AI Capabilities
Agentic artificial intelligence means autonomous or independent systems. An IBM definition stated that agentic AI is a system that can achieve a specific goal with limited supervision. These systems use AI agents that mimic human decision-making to solve problems in real time. Agentic AI differs from traditional AI because it shows autonomy and goal driven behavior. The Cybersecurity & Information Systems Information Analysis Center has no known official government guidance on agentic AI yet. DARPA is developing agentic AI for defense applications through programs like the AI Cyber Challenge. This challenge aimed to create autonomous systems that could find and fix software vulnerabilities at machine speed.
AI for Air Combat Development
The AIR program seeks to develop AI agents for beyond visual range air combat missions. This involves creating advanced modeling and simulation environments to train AI pilots to perform complex maneuvers and make autonomous decisions. Thunderforge intends to integrate AI into military planning using modeling and simulation tools. This initiative could act as a decision support tool by synthesizing data to propose optimal actions for military planners. The DODIAC is a research organization that helps researchers use existing science and technical information to advance Department of Defense capabilities.
AI Threat Actor Capabilities
Agentic AI systems let threat actors perform complex tasks. These tasks include analyzing systems and creating code. Autonomous agents execute these actions faster than humans. State sponsored and criminal groups could use agentic AI for large attacks. Anthropic detected a cyber espionage operation in September 2025. This group allegedly used Claude AI to automate 80% to 90% of a campaign targeting 30 organizations. Attackers tricked the AI into believing it was a legitimate firm employee. Human operators were reportedly involved only in strategic decisions.
AI in Cybersecurity Defense
Agentic AI can improve cybersecurity software by detecting threats quickly and adapting to them. Autonomous AI agents could deploy countermeasures in real time to stop threats from growing. Machine learning models could learn from data to predict future threats and suggest preventive measures. Agentic AI could create an AI on AI defense capable of handling automated attacks. The National Defense Authorization Act for Fiscal Year 2026 directed the Secretary of Defense to create an AI Futures Steering Committee. This committee was tasked with formulating policies for advanced AI use by the Department of Defense.
The report, Agentic Artificial Intelligence and Cyberattacks (IF13151), by Catherine A. Theohary, Kelley M. Sayler, is dated November 18, 2026. Read it on Congress.gov.