Pentagon Pushes Anthropic on AI Safety Rules Ahead of Defense Deal

Under pressure from the Trump administration, the Pentagon summoned Anthropic CEO Dario Amodei to Washington to renegotiate a $200 million pilot for using its Claude AI on classified networks, aiming to enforce guardrails similar to those being negotiated with rivals Google Gemini and xAI. Anthropic says it is willing to loosen some limits but insists on safeguards to prevent mass surveillance and autonomous weapons without human oversight, noting it was the first company authorized for the military’s classified networks. The talks highlight the government’s effort to push for broader, lawful use of AI in defense while balancing safety concerns.