The Biden administration has rejected Anthropic's proposal to broaden access to its powerful AI model, Mythos, to roughly 70 additional companies, according to a report from the Wall Street Journal. The decision stems from concerns over limited compute resources and national security risks.
Mythos is designed to identify and exploit software vulnerabilities, making it a double-edged sword. Currently, around 50 organizations—including critical infrastructure operators and government agencies like the NSA—have access through Project Glasswing. Expanding beyond that, the White House fears, could strain the computing power Anthropic has available, potentially reducing the government's own ability to use the model.
AI advisor David Sacks and other officials noted that Anthropic has less compute capacity than rivals like OpenAI and Google DeepMind. While Anthropic has inked major deals with Amazon, Google, and Broadcom to boost its infrastructure, those investments won't yield immediate results. The administration is also navigating how to continue collaborating with Anthropic after the Pentagon classified the company as a supply chain risk, a move that sparked legal threats from CEO Dario Amodei.
The standoff highlights growing tensions between AI safety, national security, and the race for computational supremacy.