US government audit outlines 16 policy options for AI cybersecurity safeguards
A new GAO technology assessment identifies challenges in testing AI security tools and proposes policy options covering governance, research, guidance and workforce skills.
The US Government Accountability Office released a technology assessment on 20 November 2026 examining how artificial intelligence can both increase the effectiveness of cyberattacks and improve defensive capabilities. The report identifies 16 policy options organised under four goals, rather than recommending a single mandatory regulatory model.
According to the assessment, AI systems can help malicious actors produce more convincing phishing content, create deepfakes and automate stages of cyberattacks. The report describes examples from 2026, including a testing environment using multiple AI agents to refine malware techniques and an attack in which stolen credentials were used against poorly secured devices.
For defenders, AI can assist with analysing network traffic, identifying vulnerabilities in source code and processing large volumes of security information. But experts consulted by the watchdog identified gaps in evaluation standards, data availability, privacy safeguards, governance and skills. The report says there are no widely accepted criteria for measuring the offensive or defensive capabilities of generative AI tools.
The options include expanded testing environments, improved lifecycle governance and stronger workforce expertise. The GAO describes these as alternatives for policymakers to consider, not rules that have already been adopted. Its findings highlight the importance of testing claims about AI security performance and maintaining human oversight where automated actions could have consequential effects.
Spotted an error? Request a correction · Our editorial principles · Contact the newsroom