TL;DR
The U.S. government has mandated a classified benchmarking process for AI models, due by August 1, affecting how AI capabilities are assessed and regulated. Participation is voluntary but may influence federal procurement preferences.
Washington has mandated a classified benchmarking process for advanced AI models, due by August 1, 2026. This process will evaluate the cyber capabilities of AI systems and designate certain models as covered frontier models. The initiative marks a significant shift toward secret standards in AI regulation, involving agencies like the NSA, Treasury, and CISA, and signals increased government oversight of AI security.
The executive order, signed on June 2, directs the Treasury, NSA, and CISA to establish a classified benchmarking framework to measure AI models’ cyber capabilities. The process will determine which models qualify as covered frontier models, with the NSA Director making the designation decisions. Alongside, a voluntary pre-release access framework allows developers to share AI models with the federal government for up to 30 days before public deployment, although participation is strictly opt-in.
This framework also includes the creation of an AI cybersecurity clearinghouse under the Treasury to facilitate information sharing on vulnerabilities between AI firms and critical infrastructure operators. The order emphasizes increased funding and hiring for AI vulnerability detection tools and federal cyber talent. However, the benchmarks will be classified, meaning developers will not see the criteria used for designation, raising concerns about transparency and oversight.
Implications of Secret AI Cybersecurity Benchmarks
This development indicates a substantial shift toward secretive regulation of AI capabilities, with classified benchmarks potentially influencing market access and government procurement. It reflects a move to treat AI cyber capabilities similarly to other weapons-adjacent technologies, where classified assessments prevent public scrutiny. For developers, opting into the voluntary framework could confer trusted partner status, impacting their ability to secure federal contracts. The move also signals increased government concern over AI security risks and the potential for future mandatory testing requirements.
As an affiliate, we earn on qualifying purchases.
Background of U.S. AI Regulatory Shifts
President Trump’s Executive Order 14409 on June 2, 2026, formalizes efforts to evaluate AI models’ cyber capabilities, following earlier moves where the government intervened to suspend certain frontier AI models with advanced cyber features. The order is a second attempt after an earlier version was reportedly pulled due to concerns over competitiveness. Historically, the U.S. has prioritized voluntary cooperation over mandatory regulation in AI governance, but this order signals a more assertive posture, with agencies like NSA and Treasury taking central oversight roles for the first time in recent months.
The European Union’s approach, exemplified by the AI Act, relies on public, contestable thresholds based on technical metrics like FLOPs, contrasting sharply with the U.S. move toward classified benchmarks.
“The classified benchmarks are designed to protect national security while enabling us to assess AI cyber capabilities effectively.”
— NSA official (anonymous)
What Details About the Benchmarks Remain Unknown
It is not yet clear what specific criteria or thresholds will be used in the classified benchmarks, nor how these benchmarks will evolve over time. The process for designating models as covered frontier models remains opaque, and the impact on AI developers who do not participate is uncertain. Additionally, the extent to which the voluntary framework will influence federal procurement decisions remains to be seen, as the framework is still being implemented.
Next Steps in Implementing and Challenging the Framework
Leading up to August 1, 2026, agencies will finalize the classified benchmarking process and establish procedures for model designation. Developers and industry groups are likely to scrutinize the framework, potentially advocating for more transparency or contesting designations. Congress may debate whether the voluntary engagement should become mandatory, or whether the benchmarks should be made public. The immediate focus will be on how the government enforces and integrates these standards into AI development and procurement practices.
Key Questions
Will the classified benchmarks be accessible to AI developers?
No, the benchmarks will be classified, and developers will not see the specific criteria used for designation, raising concerns about transparency.
Does participation in the voluntary framework guarantee government contracts?
Participation may confer trusted partner status, which could influence federal procurement preferences, but it does not guarantee contracts.
Could the benchmarks become mandatory in the future?
Yes, Congress is considering whether to shift from voluntary engagement to mandatory pre-release testing, which could formalize these standards.
How does this U.S. approach compare to Europe’s AI regulation?
The EU’s AI Act relies on public, contestable thresholds, while the U.S. is moving toward secret, classified benchmarks, representing contrasting regulatory philosophies.
What are the implications for AI innovation and competitiveness?
The move toward classified benchmarks may impact transparency and collaboration, potentially affecting U.S. competitiveness if it limits open testing and verification.
Source: ThorstenMeyerAI.com
