Flagged as a Pentagon Supply Chain Risk, Anthropic Is Probably Getting the NSA Contract Anyway
The US government has officially classified Anthropic as a supply chain risk. It's also about to give the company a contract to supply AI models to the NSA. Nobody seems particularly bothered by this contradiction.
White House Chief of Staff Susie Wiles signed off on the arrangement personally, according to the New York Times. The logic, such as it is, comes down to chips. Intelligence agencies don't have enough of Nvidia's Grace Blackwell hardware to run the latest models from OpenAI and similar providers. Anthropic's upcoming 'Mythos' model apparently runs on older silicon, which makes it the only realistic near-term option for the NSA's classified networks.
So: security concerns, meet procurement deadlines. Procurement deadlines win.
The Pentagon's original objection to Anthropic centred on the company's refusal to release its technology for 'any lawful use' — a clause that spooked defence officials worried about operational flexibility. That language has reportedly been dropped from the current deal. There's also a clause prohibiting the model from processing data belonging to American citizens, which should at least keep the lawyers busy.
The White House is apparently treating this contract as a blueprint for future government AI deals, which is either reassuring or alarming depending on how you feel about policy being drafted around chip availability.
On the hardware shortage itself, the administration has approved $9 billion for new AI chips as a longer-term fix. That still needs congressional approval, so don't hold your breath.