Pentagon: Towards Autonomous AI After Clash with Anthropic

The Pentagon Turns to Its Own AI Models

The United States Department of Defense has made a radical decision following Anthropic's refusal to provide unrestricted access to its artificial intelligence Claude for military uses. This refusal has prompted the Pentagon to develop its own AI models, despite the presence of other technology partners such as OpenAI and xAI. Cameron Stanley, the Pentagon's Chief Digital and Artificial Intelligence Officer, confirmed this initiative to Bloomberg, emphasizing the importance of this new strategic direction.

Let’s recall the episode with Anthropic. The company had refused to lift restrictions on the use of its AI, particularly for mass surveillance of citizens and the control of autonomous weapons. These restrictions aimed to avoid ethical breaches that the company deemed unacceptable. In contrast, the Pentagon demanded full, unrestricted access to these technologies, including on classified networks.

In February 2026, tensions reached a peak when the Pentagon issued an ultimatum, giving Anthropic 72 hours to lift its restrictions or face the cancellation of a $200 million contract. This ultimatum also included the threat of placing Anthropic on a blacklist, prohibiting any future collaboration.

Internal Development of AI Models

In light of this refusal, the Pentagon has decided to accelerate the development of its own AI models. These new systems are designed to be integrated directly into government environments, with complete control over data and usage. Although few details have been revealed, it is likely that these projects are already well advanced, aiming to reduce the Pentagon's dependence on private actors.

The Pentagon's digital and AI chief, Cameron Stanley, remains tight-lipped about the technical details of these new models. However, it is clear that the goal is to regain total control over the capabilities and uses of AI systems, managing everything in-house, from data to usage limits.

Ongoing Collaboration with OpenAI and xAI

Despite developing its own models, the Pentagon continues to collaborate with OpenAI and xAI. An agreement has been signed with OpenAI to deploy its AI models on classified networks. This agreement includes enhanced ethical safeguards, such as the exclusion of uses by the NSA without contractual modification and the presence of company engineers to oversee operations.

xAI, led by Elon Musk, has also integrated its Grok model into Pentagon systems, backed by a $200 million contract. These collaborations demonstrate that the Pentagon aims to diversify its partnerships to avoid reliance on a single supplier.

Implications for Anthropic

The break with the Pentagon has significant consequences for Anthropic. Defense Secretary Pete Hegseth has classified the company as a risk to the supply chain, thereby blocking any collaboration with Pentagon partners, including giants like Lockheed Martin and Boeing. Even for civilian or unclassified uses, doors are closing.

Anthropic has responded by filing in federal courts to contest this decision, citing an infringement on contractual freedom and a disputed use of the Defense Production Act by the Pentagon. The conflict now extends beyond the commercial framework, taking on a political and strategic dimension, and it is unlikely that this matter will end here.