1 min read

OpenAI Employees Allege DOD Used Azure-Hosted OpenAI Models Before Military Ban Was Lifted in 2024

Canopy Policy

Wired reports that OpenAI employees claim the US Department of Defense was testing Microsoft's Azure-hosted versions of OpenAI language models before OpenAI publicly lifted its blanket prohibition on military use in January 2024. The disclosure suggests the DoD may have accessed OpenAI capabilities through Microsoft's commercial Azure OpenAI Service as a technical workaround to OpenAI's own military use policy, without OpenAI explicitly authorising those applications. The revelation adds complexity to the ongoing dispute between Anthropic and the Pentagon over AI supply chain risk designations, and raises questions about how AI usage policies are enforced when models are deployed via third-party cloud platforms.

Key Takeaways

  • OpenAI employees allege DoD was testing Microsoft Azure-hosted OpenAI models before OpenAI lifted its military use ban in January 2024 — reported by Wired citing unnamed OpenAI sources
  • Mechanism: DoD accessed OpenAI models via Microsoft Azure OpenAI Service (a commercial service), which may not have been explicitly covered by OpenAI's military prohibition
  • Raises a key policy question for all AI vendors: does a model usage policy apply when the model is deployed via a commercial cloud partner's API tier, not directly?

Original source: Wired