Anthropic CEO warns AI lacks judgment, rejects autonomous weapons contracts

In a decisive break from industry norms, Anthropic CEO Dario Amodei has formally declined contracts with the U.S. Department of Defense for fully autonomous weapons, asserting that current AI lacks the judgment required for lethal decision-making. The stance, while consistent with Anthropic’s founding principles and its responsible scaling policy, represents a high-stakes gambit in a market expected to reach $60 billion for AI cybersecurity by 2028. The move is not merely principled—it is a strategic signal to enterprise clients that safety and control are the company's core differentiators.

Anthropic’s refusal stands in stark contrast to the posture of its competitors. OpenAI, despite a policy against deploying AI for weapons, does not publicly disclose specific contract refusals and has a partnership with Microsoft that may involve military-related cloud services. Google DeepMind has pledged not to develop lethal autonomous weapons, but Google’s history with Project Maven—a 2018 initiative that drew massive employee backlash—shows the tension between corporate ethics and government contracts. Meta’s open-source Llama models can be freely adapted by third parties for military use, placing the burden of ethical constraints on end users rather than the developer. Anthropic, by contrast, controls deployment through API licensing and contractual terms, enabling a hard line that rivals cannot easily replicate.

The implications extend beyond defense. Anthropic’s reported claim that its (unverified) vulnerability scanner, referred to internally as “Mythos,” has identified tens of thousands of software flaws and will remain unmatched for 6–12 months raises significant questions about the feasibility of such a timeline. Industry data shows that rival models, including OpenAI’s Codex and Google’s Gemini, are improving rapidly; the six-month window may be optimistic. Moreover, the ethical rejection of autonomous weapons does not preclude Anthropic from supporting non-lethal military applications, such as intelligence analysis or logistics, creating a nuanced position that some critics may view as inconsistent. The company’s true test will be whether it can maintain its principles while scaling revenue to justify its $15–30 billion valuation.

Ultimately, Anthropic’s move redefines AI ethics as a market-driven asset. By publicly setting boundaries that competitors avoid, it attracts enterprises—particularly in regulated industries—that prioritize responsible AI. The cost is real: forfeiting direct defense contracts and possibly slowing adoption in government sectors. But the payoff may be a lasting brand premium in a world where trust is the scarcest resource. The jury remains out on whether the “Mythos” claims hold water, but the strategic direction is clear: invest in safety as a competitive moat, and let rivals fight over contracts that compromise core values.