Five Unresolved Issues in OpenAI’s Deal With the Department of Defense

The end of February brought a pair of striking developments regarding military use of AI. The Department of Defense announced it would designate Anthropic as a “supply chain risk” (a threat it just followed through on) after the company held firm on a set of conditions for use of its technology: no use for “mass domestic surveillance” or “fully autonomous weapons.”

In parallel, OpenAI announced an agreement with the Defense Department for the use of its own AI systems. In an effort to assuage concerns (one that at least initially seemed to fail), the company released a statement outlining three “red lines” it had instituted for use of its systems: 1) No use for mass domestic surveillance; 2) No use to direct autonomous weapons systems; 3) No use for high-stakes automated decisions.

Seeing these limits trumpeted as key parts of a successful OpenAI deal was perplexing, as the first two appear to be exactly the same demands that caused the planned Anthropic agreement not only to fail, but to explode in shocking fashion. OpenAI has acknowledged this, stating, “our contract provides better guarantees and more responsible safeguards than earlier agreements, including Anthropic’s original contract. We think our red lines are more enforceable here …. We don’t know why Anthropic could not reach this deal, and we hope that they and more labs will consider it.”