OpenAI’s head of robotics and hardware, Caitlin Kalinowski, resigned today in protest following the company’s decision to sign a classified deal with the Pentagon.

Kalinowski, a longtime advocate for responsible AI development and former hardware leader at Oculus and Anduril, cited ethical concerns over the deployment of autonomous systems in military contexts. Her departure comes just days after the Pentagon formally designated rival Anthropic a “supply chain risk,” clearing the path for OpenAI to take on expanded defense work.

Internal sources say Kalinowski had repeatedly pushed back against military contracts that could involve autonomous targeting, surveillance agents, or robotics platforms without clear human oversight. Her exit marks the most senior public resignation at OpenAI tied directly to defense partnerships and intensifies the growing rift inside frontier AI labs over the weaponization of autonomous technology.

What Changed

The classified agreement reportedly gives the Pentagon access to advanced OpenAI models for defense applications, including potential integration with robotics and agentic systems. This follows OpenAI’s earlier restrictions on military use and stands in contrast to Anthropic’s refusal to lift its own red lines on mass surveillance and lethal autonomous weapons.

Kalinowski’s resignation letter (portions of which were shared internally) emphasized that “autonomous agents capable of real-world action should not cross into lethal or coercive domains without rigorous public oversight.”

The Bigger Picture

This is the clearest sign yet that the rapid shift toward production-grade autonomous agents is forcing hard choices at the world’s leading AI labs. As models gain native computer-use, robotics integration, and persistent memory, the line between civilian tools and military systems is vanishing.

While OpenAI moves aggressively into defense to maintain competitive edge and funding, the internal cost is mounting — talent attrition, public backlash, and growing pressure from employees who joined to build beneficial AGI, not battlefield autonomy.

The resignation lands amid a broader wave of scrutiny: Anthropic’s impending lawsuit against the Pentagon, congressional calls for oversight, and Silicon Valley debates over whether frontier labs can remain neutral in the global AI arms race.

For the autonomous AI industry, Kalinowski’s exit is a warning: the infrastructure for reliable agents is ready — but the ethical and governance framework for where those agents are allowed to operate is not.