Anthropic and the US Senate: Legislative Battle Over Military AI

A Bill to Regulate the Use of Military AI

Senator Adam Schiff, representing California, is working on a new bill aimed at "codifying" the red lines established by Anthropic regarding the use of artificial intelligence (AI) in the military domain. The primary goal of this legislation is to ensure that crucial decisions, particularly those involving life and death, remain under human control. This legislative initiative comes at a time when the question of AI usage in sensitive areas is increasingly being debated.

At the same time, Senator Elissa Slotkin from Michigan recently introduced a similar bill known as the AI Guardrails Act. This legislation aims to limit the Department of Defense's ability to use AI for mass surveillance of American citizens. It is an attempt to strengthen protections against the use of AI in contexts that could infringe on civil liberties.

Conflict with the Trump Administration

The Trump administration recently placed Anthropic on a blacklist, designating the company as a risk to the supply chain. This decision was made after Anthropic imposed limits on the use of its AI models by the military. In response, Anthropic filed a lawsuit, accusing the government of violating its constitutional rights. The company is currently awaiting a court's decision on whether it will intervene to block this designation.

Anthropic firmly opposes the use of its technologies for fully autonomous weapons and for mass surveillance within the country. This position contrasts with that of its main competitor, OpenAI, which has signed an agreement with the Pentagon. Senator Schiff expressed concern that the Pentagon's actions against Anthropic could delay America's leadership in AI.

The Stakes of the Legislation

Schiff's office is finalizing the details of the legislation, which aims to prevent the use of AI for "illicit purposes." Slotkin's bill, on the other hand, restricts the use of AI for actions such as launching nuclear weapons or surveilling American citizens. However, it also includes exceptions for "extraordinary circumstances" that require the use of AI to deploy lethal autonomous weapons.

In the bill that Schiff is drafting, the specifics regarding what constitutes an autonomous weapon or domestic surveillance are still under discussion. Schiff stated that individuals legally present in the country deserve protection, and this could extend beyond mere questions of human rights.

The Importance of Human Oversight

A fundamental principle guiding this legislative effort is the idea that there must always be a "human in the loop." Schiff explained that whenever a technology has the capacity to take a human life, there must be a human operator in the chain of command. "We do not want to delegate this type of responsibility for life and death to an algorithm," he stated.

However, this does not mean there is no role for AI on the battlefield. Schiff emphasized that AI can provide valuable information to human operators, enabling them to make informed decisions to defend the country or adjust their actions based on what AI can see in real-time on the battlefield. "The applications are very significant," he added. "They can be very beneficial from a national security and defense perspective. But they can also mean life or death, making the distinction between a civilian target and a military target, or getting those things wrong."

Political Challenges and Outlook

With a Democratic minority in both chambers, the short-term success of the bill may depend on the willingness of Republicans to be seen as critical of the administration. The upcoming midterm elections further complicate the situation, although the balance of power in Congress could change if Democrats regain one or both chambers. Schiff hopes to use legislative vehicles like the National Defense Authorization Act (NDAA) to advance his proposal.

"There is certainly bipartisan support among the public for this type of limitation," Schiff stated. "As always, you face the question when you take steps to prevent any administrative abuse, which raises issues with some of my colleagues about whether this could be interpreted as an implicit criticism of the administration. We will have to manage that, but I hope we can make it bipartisan."

Since Anthropic began its battle with the Pentagon, OpenAI has had to defend its reasons for signing terms that have sparked public backlash. Even though OpenAI claims it will insist on the same terms, Schiff stated that he would prefer not to have to rely on the Pentagon or a CEO. "I would have much more confidence, frankly, if these requirements were statutory, rather than relying on the legality of the Pentagon or the word of an AI CEO," he said.