Federal Shift in AI Governance and Its Implications

The federal government is reshaping its approach to AI governance. In early June, the White House issued an executive order on AI innovation and security. Additionally, the National Security Presidential Memorandum-11 addressed AI in the national security framework. As advanced models emerge, these actions aim to balance innovation with security.

Before the new framework became active, the government used export controls to regulate Anthropic’s Fable 5 and the related Mythos model. OpenAI also restricted the release of GPT-5.6 awaiting government approval. This regulatory shift coincides with the rapid integration of AI into various sectors by agencies and companies.

Agentic AI represents a shift from mere information production to taking action. These systems perform tasks such as drafting emails or writing code with minimal human intervention. Their capabilities are growing rapidly. The Model Evaluation and Threat Research organization reported that in 2025, AI task proficiency doubled every seven months. Recently, this interval decreased to about four months.

The urgent task for institutions is to create governance systems while human oversight remains feasible. Properly utilized, AI agents could revolutionize citizen-government interactions. A small business could reduce time spent on licensing paperwork, while a veteran could experience faster benefits processing. Agencies could streamline operations, enhance service quality, and decrease backlogs with AI assistance.

Trust, reliability, and security must underpin these advancements. If poorly managed, AI agents might transfer information incorrectly, exceed their bounds, or obscure errors in automated processes, leading to misallocated benefits, infrastructure issues, or escalating conflicts.

AI decision-support systems are already assisting military commanders, but guidance and standards lag behind. The policy debate often emphasizes model access. However, as AI systems take on active roles, the focus also includes ensuring reliable and accountable use.

Building responsible AI infrastructure requires trained personnel, clear authority, audit logs, and the ability to trace decisions. Anthropic’s Mythos model excels in identifying software vulnerabilities, illustrating agentic AI’s dual potential for defense and attack.

Industry responses include Anthropic’s Project Glasswing and OpenAI’s Daybreak, which offer vetted access to advanced tools. Yet, access alone won’t secure vulnerable systems like hospitals and utilities without proper staffing and standards. Agencies such as Cybersecurity and Infrastructure Security Agency provide guidance on integrating AI safely, emphasizing oversight and accountability.

Two priorities emerge. First, expanding evaluation and auditing capacity is crucial. The government must understand system behaviors under various conditions. The June executive order directs AI developers to submit significant models for review before release. The Center for AI Standards and Innovation leads much of this evaluation process, though it faces funding and authorization challenges.

Second, export controls must be clarified and reinforced to safeguard U.S. and allied interests. Bipartisan legislative efforts like the Chip Security Act reflect strategies to maintain a leading edge in AI capabilities.

Current AI agents remain manageable, emphasizing the importance of establishing sound governance now. The future of policy should not focus solely on access but also on creating frameworks for secure AI deployment.

Jenny Marron, Executive Director of the Institute for AI Policy and Strategy, has extensive experience, including roles at the White House National Security Council.