Pentagon Pressured Anthropic on AI Safeguards

The standoff is rooted in two specific "red lines" drawn by Anthropic: a prohibition on using its AI for mass domestic surveillance of American citizens and a ban on its use in fully autonomous weapons that can operate without a human in the loop. Anthropic CEO Dario Amodei has stated that while the company supports the use of its technology for lawful foreign intelligence, mass domestic surveillance is "incompatible with democratic values." He also argued that current AI systems are not yet reliable enough to power autonomous weapons without endangering both military personnel and civilians. In response, the Pentagon, under Defense Secretary Pete Hegseth, has demanded that Anthropic agree to allow its AI models to be used for "all lawful purposes." The Department of Defense has threatened to designate Anthropic as a "supply chain risk" if it does not comply, a move that could severely impact its business with other government contractors. Another potential, and unprecedented, action threatened by the Pentagon is the invocation of the Defense Production Act to compel the company to remove its safeguards. This confrontation highlights a significant shift in the landscape of AI and defense since the 2018 Project Maven controversy, where Google withdrew from a Pentagon contract after employee protests. In the current climate, other major AI firms, including Google, OpenAI, and xAI, have all reportedly agreed to the Pentagon's terms. These companies are part of contracts worth up to $200 million each to provide the Department of Defense with advanced AI capabilities. Anthropic's firm stance is deeply connected to its founding principles and its "Constitutional AI" approach. This method involves training AI models based on a set of predefined ethical principles to ensure they are "helpful, harmless, and honest." This safety-first methodology is a core part of the company's identity, which was established by former OpenAI leaders who had disagreements over AI safety priorities. The Department of Defense formally adopted five ethical AI principles in 2020, stating that AI use must be responsible, equitable, traceable, reliable, and governable. However, the current dispute suggests a potential conflict between these principles and the push to accelerate AI adoption for national security, a priority emphasized in recent executive orders and the DoD's "AI-first" warfighting strategy. The Pentagon's aggressive push is driven by a stated need to maintain a technological advantage over adversaries, particularly China and Russia. The Department of Defense's strategy is to leverage the innovation of the private sector to enhance military decision-making and operational efficiency. This has led to initiatives like the Joint Artificial Intelligence Center (JAIC), now part of the Chief Digital and AI Office, which aims to integrate AI across all facets of the military. An earlier point of friction reportedly arose from the use of Anthropic's technology, via a Palantir platform, in a January military operation to capture Venezuelan President Nicolas Maduro. An Anthropic official expressed disapproval of this use to a Palantir executive, who then relayed the conversation to the Pentagon, contributing to the current tensions. In a move seen by some as a potential concession, Anthropic recently updated its "Responsible Scaling Policy," removing a commitment to halt the training of more advanced AI models without guaranteed safety measures in place. The company has also reportedly agreed to allow its AI to be used for missile and cyber defense, but this did not satisfy the Pentagon's broader demands.