AI specialist Anthropic still calls the Pentagon’s final contract proposal unacceptable and does not plan to completely scrap its policy on safe AI use.
Anthropic CEO Dario Amodei states that “virtually no progress” is being made in negotiations with the Pentagon regarding the continued use of the AI tool Claude, according to Axios. A final contract proposal from the U.S. government apparently contains no significant changes. Anthropic is digging in its heels and refuses to supply AI under unrestricted terms, even as the Pentagon threatens severe consequences.
Use without borders
The U.S. Department of Defense wants Anthropic to impose no terms of use on the LLM, which is the only one certified and used on classified systems. Anthropic refuses to comply. For Amodei, restrictions on the use of AI must remain in place.
A thorn in the side of the department is, on the one hand, a ban on building robots that can autonomously attack and kill with the help of AI, and on the other hand, a ban on using AI for domestic mass surveillance of citizens.
This is a major problem for the Pentagon under the leadership of former Fox News anchor Pete Hegseth, a staunch Trump loyalist who wants a free hand to deploy AI without restrictions. He believes only the law should set limits on the military’s use of AI, not the AI developer itself. Legislation on the subject, especially in the US, is lagging behind reality and consequently imposes few real restrictions.
Deadline with retaliation
Anthropic has until 5:01 PM local US time to agree to the Pentagon’s proposal. Hegseth is threatening severe retaliation if this does not happen. There are two possible paths, both of which are unprecedented in this situation.
On one hand, Hegseth wants to label Anthropic a ‘supply chain risk.’ A company with that label is prohibited from doing business with the Pentagon. Furthermore, other organizations worldwide that supply the Pentagon would also be barred from working with such a high-risk enterprise. This label is not intended for domestic companies at all, but rather for organizations from hostile nations, such as Chinese companies with government ties. Nevertheless, Hegseth intends to use it to pressure Anthropic.
On the other hand, the Secretary of Defense is brandishing the Defense Production Act (DPA). This is a law that allows the US, in the name of national security, to compel a company to develop a custom product. De facto: AI that is allowed to control killer robots and surveil citizens.
Amodei states that the company remains open to further discussion. Anthropic is willing to adjust its terms of use for the Pentagon but refuses to cross the aforementioned red lines.
Global consequences
The story is more than just a domestic US matter. Most AI companies at the forefront of AI development today are based in the US. With these threats and deadlines, the US government is now demonstrating how heavily it intends to influence these companies. Developing ethical AI is being prohibited, under threat of immense economic consequences or coercion.