Federal Shift in AI Governance and Its Impact on National Security

40 Views
admin
July 1, 2026
Cybersecurity Technology

Federal efforts to govern AI are changing significantly, although much remains unresolved. In June, the White House issued two key directives: an executive order on AI innovation and security, and National Security Presidential Memorandum-11, related to AI in national security. These actions follow advances in new AI models with enhanced cyber capabilities. Preceding the establishment of a comprehensive review framework, the administration used export-control authority to limit Anthropic’s Fable 5 and its underlying Mythos model. OpenAI has delayed GPT-5.6’s release pending governmental approval.

This shift occurs as agencies and businesses rapidly incorporate AI into daily operations. The broader implications concern secure deployment and the trustworthiness of these systems in performing tasks independently. Agentic AI involves delegating tasks such as drafting emails, searching databases, filing forms, coding, monitoring networks, or routing requests. These systems, now capable of multi-step actions before human intervention, continue to improve swiftly.

The Model Evaluation and Threat Research organization assesses the efficiency of advanced AI systems by comparing them to human task completion times. In 2025, the benchmark was doubling every seven months; recent estimates update this to every four months. Institutions must enhance their capacity to govern AI while human control remains viable. Properly managed, AI could revolutionize government-citizen relationships. For instance, small businesses might allocate more time to customer service rather than paperwork, and veterans could experience faster benefits claim processing. Agencies might reduce procedural redundancies and improve service quality.

However, ensuring trust, reliability, and security is crucial. Poorly governed AI might misallocate resources, disrupt infrastructures, or escalate conflicts. AI in decision-support systems provides military target suggestions and integrates into sensitive operations, yet lacks comprehensive guidance and standards.

AI policy discussions often revolve around access: determining who acquires models, chips, data, and energy. As AI begins to perform tasks, policymakers must ensure systems are used securely and accountably, with capabilities understood by governments and the public. Building a responsible usage framework requires trained personnel, clear authority, auditing capabilities, and post-decision reconstruction.

Cybersecurity illustrates the stakes. Anthropic’s Mythos model excels at identifying software vulnerabilities, showing rapid advancement potential for defensive and offensive actions. In response, initiatives such as Anthropic’s Project Glasswing and OpenAI’s Daybreak are providing vetted defenders with advanced tools under a differential access strategy.

The White House focuses on access, yet lacking staff, standards, and integration could leave vital services vulnerable. Agencies such as the Cybersecurity and Infrastructure Security Agency, National Security Agency, and international partners stress careful AI integration, emphasizing permissions, monitoring, and human oversight.

Policy emphasis should be on evaluation and auditing capacity development. For government reliance on AI, understanding how systems operate, maintain boundaries, and handle cyber-relevant tasks is essential. While a June executive order initiated a model review process, execution is progressing. Evaluation science must keep up, as models knowing they are being evaluated complicate performance metrics.

The Center for AI Standards and Innovation drives this evaluation, supported by collaborations with OpenAI and Anthropic. Despite progress, the Center’s current funding and lack of Congressional support limit its potential, highlighting the need for additional resources to enhance its national security deployment role.

Strengthening export controls is another priority. It is crucial not only to possess the most capable models but to shape agentic system development based on valued institutions and ethics. Bipartisan initiatives such as the Chip Security and Stop Stealing Our Chips Acts aim to maintain U.S. leadership.

Current AI systems remain manageable. Current deployments set the stage for institutional practices as agentic systems advance. Future AI policies should prioritize creating governance frameworks for dependable, accountable, and secure utilization.

Jenny Marron is Executive Director of the Institute for AI Policy and Strategy. She previously served at the White House National Security Council and U.S. Department of State.