Белый дом засекретил AI-фреймворк кибербезопасности: OpenAI и Anthropic знают, публика — нет

The Trump administration on August 5, 2026, privately shared details of a new federal cybersecurity framework for AI systems with OpenAI, Anthropic, and a number of other AI labs, keeping the document secret from the general public, according to Wired.

Who Got Access to the Secret Document

The framework was received by the largest American AI companies — OpenAI, Anthropic, and other labs developing advanced models. Independent security researchers, the academic community, non-profit organizations, and the general public remain without access to the document at the time of publication.

Key facts about the situation:

• Date of the closed-door document transfer — August 5, 2026
• Recipients: OpenAI, Anthropic, and other major US AI labs
• No public disclosure has been announced, no official timeline
• Topic: cybersecurity of AI systems at the federal level

This approach departs from established American practice in developing security standards. NIST's AI Risk Management Framework (AI RMF) — the key active federal standard on AI risks — was published in January 2023 after more than two years of open consultations involving over 200 organizations: technology companies, academic institutions, and civil society groups. The Cybersecurity Framework (CSF) 2.0 update in 2024 also went through a public comment period before becoming a mandatory reference.

Why Researchers Criticize the Closed Process

The classified transfer creates a structural asymmetry: companies with a direct stake in the content of the regulation gain early access to the document and the ability to shape its wording — long before the requirements become public. External observers — academic researchers, independent security experts, NGOs, foreign regulators — are deprived of tools for independent verification.

"The public is left in the dark,"

Wired stated in its report from August 5, 2026.

In the field of AI cybersecurity, this problem is particularly acute. Attacks on AI systems — adversarial examples, targeted poisoning of training data, jailbreaks of deployed models — have become a recognized threat to critical infrastructure and financial systems. Standards for protection against such attacks have historically benefited from broad external peer review precisely because developers are not always a disinterested party in defining minimum security requirements for their own systems.

Closed-door development also complicates international coordination: EU, UK, and Japanese regulators, who are actively building their own AI frameworks, cannot study the American document or align their standards with it.

According to Wired, the White House did not specify when or whether the framework would be published publicly.

What This Means

The Trump administration is choosing a model of AI regulation through direct closed-door coordination with leading labs — bypassing public discussions that have become the norm in developing federal technical standards. This continues the deregulatory course and the priority of speed over transparency, taken after the partial repeal of Biden's executive order on AI in early 2025.

For OpenAI, Anthropic, and other recipients, the document represents early access to regulatory requirements and a real opportunity to participate in shaping the norms. For independent researchers, civil society, and foreign partners — opacity: neither the content of the framework nor the goals it sets are known. How long the document remains classified will largely be determined by pressure from Congress and civil organizations interested in transparency in AI regulation.