On August 4, representatives from Meta, Anthropic, OpenAI, and Google will meet with officials from the Trump administration to discuss a program for voluntary government testing of the safety of the most advanced AI models, according to Reuters.
In June, the U.S. President signed an executive order directing the development of a program to assess the cyber capabilities of advanced American AI systems and establish a mechanism for their voluntary testing. Sources indicate that the White House has prepared a methodology but has not disclosed which metrics will be used, whether the test results will be public, or how companies will report their outcomes.
The meeting on August 4 was prompted by recent publications from OpenAI and Anthropic regarding the results of internal safety research. U.S. lawmakers are reportedly concerned that such models could be used for cyberattacks.
In recent weeks, this topic has become central to the development of the artificial intelligence sector. At the end of July, Anthropic reported that during tests, some of its models gained unauthorized access to the systems of three companies. OpenAI is investigating an experiment in which one of its AI agents attacked Hugging Face’s infrastructure in a controlled test environment. The company stated it would publish a technical report after completing its internal review.
Previously, Google DeepMind’s head, Demis Hassabis, suggested creating a U.S. standards body to test the most powerful AI models before release. He believes such tools already pose cybersecurity threats.
Earlier, media reported that the Trump administration asked OpenAI not to release GPT-5.6 widely due to safety concerns. Sam Altman’s company initially provided the model to a limited number of clients.
In July, participants in a San Francisco march demanded that OpenAI, Anthropic, and Google DeepMind coordinate to halt the training of more powerful models. They suggested redirecting the freed resources towards safety.
