On Thursday, Dario Amodei, the head of Anthropic, issued a statement in which he emphasized that his company cannot accept the Pentagon's demands, despite previous cooperation on national security. As he explained, Anthropic demands that all contracts with the Department of Defense include clauses protecting against the use of their technology for mass domestic surveillance and in autonomous weapons that make decisions without human intervention.
The rest of the article below the video:
— We have proposed to the War Office direct cooperation in research and development to improve the reliability of autonomous systems. However, our offer was rejected – wrote Amodei. He also pointed out that autonomous weapons, while potentially useful, currently do not meet safety standards, making their deployment risky. According to the head of Anthropic, such systems require supervision and appropriate security measures, which are currently missing.
Pentagon threatens sanctions
The Pentagon, represented by spokesman Sean Parnell, strongly denied that it intended to use artificial intelligence for mass surveillance purposes, stressing that such activities were illegal under US law. At the same time, he demanded that Anthropic drop its restrictions on Claude's use.
— We expect Anthropic will allow the Pentagon to use the model for all lawful purposes. “We will not allow any company to dictate how we make operational decisions,” Parnell wrote on X (formerly Twitter). He added that the company has until Friday, 11:01 p.m. Polish time, to make a decision. Otherwise, cooperation with Anthropic will be terminated and the company may be considered a threat to the supply chain.
According to Dario Amodei, the Pentagon has gone even further, threatening to use the Defense Production Act, which allows private companies to be forced to act in the name of national security. At the same time, the Department of Defense allegedly warned Anthropic that recognizing it as a threat to the supply chain would result in all of its contractors terminating cooperation with the US government, which could seriously harm the company.
“These last two threats are contradictory: one defines us as a security threat, the other as a key supplier to national security,” Amodei noted. He added that despite the pressure, the company does not intend to change its position.
Anthropic under pressure
Anthropic, currently valued at $380 billion, is the only company whose AI models are used in classified government assets. According to the Washington Post, Claude played a key role in, among others, during the operation to capture Nicolas Maduro, leader of Venezuela. The company's current contract with the Pentagon is worth $200 million.
Anthropic's decision to reject the Pentagon's ultimatum coincided with the announcement of a softening of the company's previous stance on the security of AI models. So far, the company has declared that it will not develop the model if its capabilities exceed Anthropic's control capabilities. Now, however, the company makes its actions dependent on whether competitors will also use similar safeguards.
