Tags: US government to test AI before inauguration

International Desk: The world’s leading technology companies have agreed to an unprecedented deal to avoid risks from artificial intelligence, or AI, technology. From now on, Google, Microsoft and Elon Musk’s XAI will allow the US government to test any of their powerful AI models before they are released to the public.
The US Commerce Department confirmed on Tuesday that OpenAI and Anthropic had also been added to the process, bringing the five major US “frontier AI labs” under government oversight.
The Center for AI Standards and Innovation (CAISI), under the U.S. Department of Commerce, will conduct the assessment. The risks that will be assessed by a new AI system include cybersecurity, i.e. the misuse of AI in hacking or digital attacks; biosecurity, i.e. the potential for AI to help create chemical or biological weapons; and national security, i.e. any threat to the overall security of the country.
The Trump administration reorganized the AI Safety Institute, established during the Biden administration, in June and renamed it CAISI. Commerce Secretary Howard Lutnick said the move aims to reduce unnecessary regulation, but will not compromise on security.
The center is now leading national security assessments as part of Trump’s AI Action Plan.
While all the institutions have signed the agreement, the U.S. government’s relationship with Anthropic is somewhat complicated. The Pentagon declared the company a “supply chain risk” in March after it refused to lift restrictions on its use of AI in autonomous weapons. The company is currently in a legal battle with the government and the administration plans to stop using it for government purposes within the next six months.
While the testing process is currently “voluntary,” the Trump administration is considering making it mandatory in the future. It is rumored that an executive order could soon mandate pre-release testing for AI models.
“This industry collaboration will further expand our work in the public interest and at a critical time,” said Chris Fall, director of CAISI. So far, the organization has completed evaluations of more than 40 AI models.