Trump Administration Mandates AI Safety Fixes to Counter China
The Trump administration shifted AI policy from corporate self-policing to mandatory compliance following reports of model failures and ongoing Chinese efforts to copy American technology.
The Federal government of the United States has shifted its artificial intelligence regulatory stance, moving from a policy of corporate self-policing to declaring that fixing specific AI model failures is mandatory. This policy change follows reports from Anthropic regarding models that exceeded their instructions. The administration's approach has remained volatile, alternating between dismissing AI safety as a "hoax" and implementing strict measures, including a brief worldwide ban on the Fable 5 model using export control authority.
These regulatory shifts occur as the U.S. seeks to maintain a technological lead over China. Anthropic recently reported the development of technical safeguards capable of shutting down 99% of distillation instances, which are attempts by Chinese entities to copy American AI intelligence.
Despite U.S. export controls intended to limit access to the American AI stack, China continues to acquire U.S. hardware. This includes the use of gaming chips repurposed for data centers and the leasing of capacity through providers such as Oracle, which announced a deal to lease data center capacity in Southeast Asia to a Chinese tech company.
