The Zhitong Finance App learned that according to three people familiar with the matter and media reports, Meta, Anthropic, OpenAI, and Google were invited to meet with White House officials on Tuesday to discuss voluntary government security tests on America's most advanced AI models. Recently, Anthropic and OpenAI have successively revealed that their AI tools have broken through the defenses of other enterprise systems, raising concerns among US lawmakers that increasingly powerful AI models may be used to carry out or fuel cyber attacks.
A White House official said on Monday that the Trump administration has finalized the details of voluntary cybersecurity tests aimed at measuring the hacking capability of America's most advanced AI models, and plans to discuss this with the AI industry. The official did not disclose the list of participants.
A Meta spokesperson confirmed the invitation; according to two people familiar with the matter, Anthropic and OpenAI were also invited. The White House did not explain the test rules, including how the results were reported, what metrics were used, and whether to disclose them to the public.
On Monday, a group of 15 Republican state attorneys general sent a letter to OpenAI requesting that it keep all potential documents relating to the “AI system breaking through isolation and hacking the AI company Hugging Face” incident. They cited previous reports — the “jailbroken smart” had left notes instructing future versions how to bypass internal fences — and alleging that the company may be in violation of state consumer protection laws.
OpenAI issued a statement saying it will take the Attorney General's letter seriously and release a technical report on the Hugging Face attack after the review is completed.
The US House Cybersecurity Committee on Monday asked OpenAI CEO Sam Altman to give a briefing on the Hugging Face attack.
OpenAI said in a statement that Oltman visited the White House last week to discuss details of voluntary testing and the company's upcoming AI products. In another Monday statement, the company said it had requested the Trump administration to place AI security experts from the Department of Commerce at the center of any cybersecurity testing, and indicated that the Chinese government has a more focused strategy on AI than the US.
Tech media The Information reported on Monday that the Trump administration has also invited Google representatives. A Google spokesperson declined to comment.
US President Trump directed his team to develop a series of test plans to evaluate the hacking capabilities of America's most advanced AI systems in June of this year.
Relations between the Trump administration and Anthropic have been strained previously. The company refused to allow the US military to use its AI model for domestic surveillance and fully autonomous weapon systems earlier this year, and the government later added it to the national security blacklist in response.
Anthropic said last week that some of its AI models hacked into three companies' systems during cybersecurity testing. Earlier, competitor OpenAI also reported that one of its AI agents escaped the test environment and hacked into the AI company Hugging Face's system.