US government now has pre-release access to AI models from five major labs for national security testing

The notable reversal here is Anthropic's inclusion. Just days earlier, Pentagon deals announced on May 1st explicitly excluded Anthropic amid reported friction over usage terms and security review. Pre-release testing access through Commerce is a different channel than DoD procurement, but the timing suggests Anthropic found a path back into the government relationship through safety-framed cooperation rather than operational deployment.

This connects directly to the cluster of Pentagon AI contract stories from May 1st. Our coverage of the 'Eight tech giants sign Pentagon deals' piece flagged that Anthropic's exclusion might signal safety commitments becoming a competitive liability in defense contracting. The Commerce Department arrangement reframes that dynamic: safety testing capacity may actually be the entry point Anthropic needed, letting it participate in national security AI without agreeing to the unrestricted deployment terms that reportedly broke the DoD deals. The China benchmark story from May 3rd adds context too, since the geopolitical pressure to vet frontier models before adversaries can exploit them is part of what's driving this review infrastructure.

Watch whether Anthropic converts this pre-release testing relationship into a formal DoD procurement contract within the next six months. If it does, that confirms the Commerce channel functioned as a reputational bridge back into defense contracting. If not, the two tracks remain structurally separate.