US Government Strikes Landmark Safety Deal with AI Giants

In a move that signals the end of the "Move Fast and Break Things" era for artificial intelligence, the US Department of Commerce has finalized agreements with Google DeepMind, Microsoft, and xAI to vet new models before they reach the public. The Center for AI Standards and Innovation (CAISI) will now serve as a federal gatekeeper for the industry’s most powerful tools.

The agreement focuses on identifying "catastrophic risks" related to cybersecurity and bioweapons. This regulatory shift comes as Anthropic prepares for the wider release of "Mythos," a model so advanced that safety experts feared its potential for autonomous exploitation of critical infrastructure. Under the new deal, companies will provide the government with early, "un-sandboxed" versions of their models for rigorous stress testing.

While the administration frames this as a win for national security, critics in the tech community worry about the slowing pace of innovation. Nevertheless, the consensus among the majors Microsoft and Google is that "large-scale public safety risks" can no longer be handled by internal corporate boards alone. For the first time, the code that will shape the future of the economy must first pass through a government lens.