Pentagon Signs AI Deals with Several Companies After Blacklisting Anthropic Over Safeguard Concerns

War stated in a January memo that it would contract only with artificial intelligence companies agreeing to any lawful use and willing to remove safeguards related to surveillance and autonomous weapons. This policy led to a conflict with Anthropic, the company behind the AI models Claude and Claude Myhtos.

Anthropic released a statement in late February supporting AI use for defending the United States and other democracies against adversaries like China, but it also highlighted potential abuses. Anthropic warned that in a narrow set of cases, the government could misuse AI for mass domestic surveillance, which it described as incompatible with democratic values.

The company expressed concerns about fully autonomous AI weapons lacking reliability without proper guardrails. This position connected to ongoing debates about the Foreign Intelligence Surveillance Act.

“We support the use of AI for lawful foreign intelligence and counterintelligence missions. But using these systems for mass domestic surveillance is incompatible with democratic values. Anthropic sued the Pentagon in March to reverse this blacklisting. These companies agreed to the military's terms for unrestricted use. The agreements reflect a broader industry shift toward competitive AI development. For instance, OpenAI, founded with principles of altruism and transparency, later focused on commercializing products like ChatGPT and increasing secrecy, according to a book by investigative reporter Karen Hao.”

Musk filed a lawsuit against OpenAI, arguing that its CEO Sam Altman and president Greg Brockman should not run a for-profit company due to trustworthiness concerns. Musk has stated that AI poses an existential threat to humanity. The lawsuit highlights changes in OpenAI's approach from its original goals of preventing AI abuse.

Karen Hao's book noted that OpenAI abandoned early commitments to transparency and democratic governance in the early 2020s, accelerating commercialization without addressing potential harms. This evolution mirrors trends in the AI sector, where initial prudence gave way to rapid advancement.

The Pentagon's decisions align with this industry dynamic, opting for partners willing to forgo certain restrictions.