Anthropic rejected Pentagon's request for unconditional military use of AI, even with warnings

Short Summary

* Situation:

Anthropic declined to grant the U.S. Department of Defense (MoD) unrestricted access to its AI models, despite a threat of losing a $200 million contract.

* Context:

During a White House meeting, Defense Secretary Pete Hegseth gave Anthropic until the end of Friday’s business day to consider the offer, but the company announced its refusal on Thursday.

* Positions of the parties:

- The MoD demands full access to AI technologies for unrestricted military use.

- Other major players (OpenAI, Google, Elon Musk’s xAI) have agreed to allow the Pentagon to use their systems “for all lawful purposes” on unclassified networks and are working on agreements for classified networks.

* Anthropic’s reaction:

In a statement by CEO Dario Amodei, it was emphasized that the company is actively collaborating with military organizations and believes in the existential importance of AI for U.S. defense. However, it has never opposed specific military operations and argues that AI can undermine democratic values only in very limited cases.

Amodei identified two “red lines”:

1) no mass surveillance of U.S. citizens;

2) a ban on the development of lethal autonomous weapons (or any weapon capable of destroying targets without human control).

The company is not yet ready to deploy fully autonomous systems.

* Potential consequences:

The Pentagon has already asked major defense contractors to assess their dependence on Claude. If the MoD deems Anthropic a “supply chain risk,” it could lead to the company being classified as a national security threat, a status typically applied to Chinese and Russian firms.

The Pentagon stated its readiness to require confirmation from all suppliers and contractors that they do not use “anthropogenic” models. If Anthropic’s services are discontinued, the company is prepared to ensure a smooth transition to another provider without disrupting military operations.

* Legislative measures:

The Pentagon is considering applying the 1950 Defense Production Act, which allows federal authorities to intervene in private companies’ activities to meet MoD requirements.

* Key message from Anthropic:

The threat from the Department does not change the company’s stance: they cannot ethically comply with requests for full access. If cooperation ends, Anthropic is ready to support a smooth transition and continue providing its models under agreed terms.

Conclusion:

Anthropic remains committed to its privacy principles and limits on military AI use, despite the risk of losing a substantial contract. The Department of Defense seeks unrestricted access to cutting‑edge technologies from all key suppliers. Possible solutions include federal intervention or switching to alternative providers.