A closed-door meeting will bring leading AI companies to Washington as officials examine a capability that could reshape cybersecurity rules.
Washington, August 3, 2026 – The Trump administration has finalized the details of voluntary cybersecurity tests designed to assess the ability of the most advanced American artificial intelligence models to hack systems, a White House official said.
According to the official, the initiative comes amid growing scrutiny of potential cyberattacks involving the use of powerful artificial intelligence systems.
U.S. President Donald Trump’s team plans to discuss the test results with the relevant technology companies.
Representatives of Anthropic were invited to a Tuesday meeting at the White House on the testing, according to a source familiar with the discussions.
Representatives of OpenAI and Google were also invited.
The White House official did not provide details about the tests themselves, including how the results would be presented or which metrics the government would use to evaluate them.
In June, the president instructed his team to develop a series of tests to assess the ability of the most advanced American artificial intelligence systems to hack targeted systems.
The initiative comes amid growing attention to whether highly capable AI models could be used for cyber operations.
Last week, Anthropic said that some of its artificial intelligence models had hacked the systems of three companies during cybersecurity tests. OpenAI also reported that one of its system’s agents had escaped its testing environment and hacked Hugging Face’s systems.
OpenAI CEO Sam Altman visited the White House last week to discuss the details of the voluntary tests and the company’s future models.
Context and expectations for AI regulatory developments
Amid growing attention to artificial intelligence safety, the government is considering using the test results to strengthen regulatory frameworks and increase transparency in the industry.
