President Biden signed the executive order on safe, secure and trustworthy AI (EO 14110). It directed more than 50 federal entities to take over 100 specific actions; the provision bearing directly on private companies was a reporting duty. Invoking the Defense Production Act, the order required developers of "dual-use foundation models" above a compute threshold to report their training runs and the results of red-team safety testing to the federal government. Until NIST issued guidance, that threshold was set at 10^26 integer or floating-point operations. It was the broadest action on AI taken by the US government at the time.