What happened
On 20 November 2026 GAO published report GAO-27-108458 assessing how AI technologies magnify the reach, precision and speed of cyberattacks while also expanding cyber defence capabilities. It documents agentic AI being used to autonomously carry out multiple attack stages with minimal human input, and notes that cyber defenders face challenges because there are no widely accepted evaluation criteria for assessing AI offensive/defensive capabilities and insufficient data and guidance on using AI tools. GAO identifies four potential policy goals (address gaps in cybersecurity guidance/practice, address AI-related research gaps, govern AI tools throughout their cybersecurity lifecycle, and enhance the cybersecurity workforce) and 16 associated policy options — including harmonizing cybersecurity requirements, expanding AI testing environments, enhancing innovation and research in agentic AI identity management, and developing governance approaches to allow safe adoption of AI for cybersecurity.
Why it matters
This is an authoritative US government baseline and policy-options document that will inform congressional and federal-agency action on AI-security governance, standards and evaluation. It provides a structured, expert-informed menu of gaps and options that standards bodies and practitioners can reference when shaping AI-security guidance, evaluation criteria and agentic-AI identity/governance work, and it underscores the current absence of widely accepted AI capability evaluation criteria.
Action needed
Practitioners and standards bodies should reference the report's policy goals and options (e.g., AI testing environments, agentic AI identity management, harmonized requirements) when shaping AI-security governance, evaluation and assurance programmes, and monitor for any subsequent congressional or agency action it triggers.