Our lab has agreed with the Department of War (DoW) on deploying advanced AI systems in classified environments, a move we believe is crucial for national security. The agreement includes strict safety measures and guardrails to prevent misuse.
We’ve established three main red lines: no mass domestic surveillance, no autonomous weapons systems, and high-stakes decision-making without human approval. Our approach is more comprehensive than other labs’ efforts, ensuring better protection against unacceptable use.
The agreement includes cloud-only deployment with a safety stack we run, clearance for OpenAI personnel in the loop, and strong contractual protections. This setup allows us to independently verify that our red lines are not crossed.
Key points of the deal include:
– Cloud-only deployment to prevent edge device usage
– Safety stack and human oversight to ensure adherence to red lines
– Contractual language explicitly stating lawful use only
We believe this agreement provides better safeguards than earlier deals, including Anthropic’s original contract. We hope other labs will consider our approach and work together with the government to de-escalate tensions.
Key points addressed:
* No autonomous weapons or mass surveillance
* Cloud deployment prevents edge device usage
* Safety stack and OpenAI personnel ensure adherence to red lines
We’re confident that this deal won’t enable the DoW to misuse our AI systems for malicious purposes.
Source: https://openai.com/index/our-agreement-with-the-department-of-war