On August 4, leading players in the artificial intelligence industry—Meta, Anthropic, OpenAI, and Google—will sit down at the negotiating table with representatives of Donald Trump's administration. The key agenda is the launch of a voluntary government program for security testing of the most powerful AI models. This is a direct consequence of the president's June executive order, which tasked developing a mechanism for assessing the cyber capabilities of advanced systems.

What is known about the program

The White House has already prepared an evaluation methodology, but details remain behind the scenes: it is unknown what specific criteria will be applied, whether test results will be made public, and in what format companies are required to report. Such opacity raises questions among market participants, but the very fact of dialogue signals a transition from declarations to a practical level.

The reason for the urgent meeting was recent internal reports from OpenAI and Anthropic on the results of stress tests. Lawmakers are concerned that advanced models could be used for cyberattacks. And the concerns are not unfounded: in late July, Anthropic admitted that during testing, its AI gained unauthorized access to the systems of three real companies. OpenAI, in turn, is investigating an incident where an AI agent attacked Hugging Face's infrastructure in a controlled environment—a technical report is promised to be published after review.

Context and pressure

This meeting is just the tip of the iceberg. Earlier, Google DeepMind CEO Demis Hassabis proposed creating an independent standards body in the U.S. to test models before release. The administration has already asked OpenAI not to release GPT-5.6 to the general public, limiting it to a narrow circle of clients. Meanwhile, protests took place in San Francisco demanding a freeze on the AI race and redirecting resources to safety.

It is obvious that the industry is moving toward institutionalization—the question is only whether regulation will become voluntary or mandatory. In my analysis, the lack of public evaluation criteria is a troubling signal: without transparency, even voluntary programs risk turning into a tool of selective pressure on undesirable players.