Our agreement with the Department of War

2026-02-27 · Source: OpenAI News · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, AI Policy & Governance · Depth: Intermediate, medium

Summary

OpenAI announced on February 28, 2026, an agreement with the U.S. Department of War (DoW) to deploy advanced AI systems in classified environments, requesting that the same terms be made available to all AI companies. This agreement incorporates three primary "red lines": no use of OpenAI technology for mass domestic surveillance, autonomous weapons systems direction, or high-stakes automated decisions like "social credit." OpenAI emphasizes that its agreement includes more guardrails than previous classified AI deployments, such as Anthropic's, by retaining full discretion over its safety stack, deploying via cloud, involving cleared OpenAI personnel, and implementing strong contractual protections. The deployment architecture is cloud-only, preventing edge device deployment that could facilitate autonomous lethal weapons, and includes independent verification mechanisms. The contract explicitly references existing laws and policies to ensure compliance, even if future regulations change.

Key takeaway

For CTOs and VPs of Engineering evaluating AI deployments in sensitive sectors, your organization should prioritize agreements that include explicit, enforceable red lines against misuse. Insist on cloud-only deployment architectures and retain control over your AI safety stack, ensuring cleared personnel are involved. This approach, exemplified by OpenAI's DoW agreement, mitigates risks associated with autonomous weapons, surveillance, and high-stakes automated decisions, providing a robust framework for responsible AI integration.

Key insights

OpenAI's DoW agreement prioritizes safety and ethical AI use through strict guardrails and cloud-only deployment.

Principles

• Retain full control over AI safety stack.
• Cloud-only deployment enhances control.
• Human oversight is critical for AI systems.

Method

OpenAI's method involves cloud-only deployment, an independently run safety stack with classifiers, cleared OpenAI personnel in the loop, and contractual language referencing current laws to enforce red lines against misuse.

In practice

• Implement cloud-only deployments for sensitive AI.
• Integrate independent verification mechanisms.
• Ensure contractual language references current laws.

Topics

• AI Deployment
• National Security AI
• AI Safety Guardrails
• Autonomous Weapons
• Government AI Collaboration

Best for: CTO, VP of Engineering/Data, Executive, Policy Maker, AI Ethicist, Director of AI/ML

Related on AIssential

• OpenAI reveals more details about its agreement with the Pentagon — OpenAI's DoD agreement emphasizes multi-layered safeguards and deployment architecture to prevent misuse, despite public scrutiny.
• Anthropic might be DONE (48 hours left) — AI safety policies face immense pressure when conflicting with national security and military demands.
• OpenAI In, Anthropic Out — AI companies are navigating complex ethical and operational dilemmas regarding military applications and government partnerships.
• Pentagon pushes AI companies to deploy unrestricted models on classified military networks — The Pentagon seeks unrestricted commercial AI access for classified military operations, facing resistance from companies like Anthropic over safety.
• Trump Administration Shuns Anthropic, Embraces OpenAI in Clash Over Guardrails — Government AI partnerships are shifting, with Anthropic facing exclusion while OpenAI and xAI gain classified access.
• Google & OpenAI employees: the government’s response to a vendor insisting on restrictions looks like punitive overreach that could chill safety debate across the entire frontier AI ecosystem. — Vendor-imposed AI safety guardrails are vital temporary measures in the absence of robust public law.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by OpenAI News.