OpenAI built its reputation on safety-first principles. For years, the company maintained a clear rule: military applications were off limits. But a newly reported discovery suggests that a significant loophole may have undermined that commitment long before OpenAI formally changed course.
According to sources speaking to Wired, the Pentagon was testing the company's models through a convenient loophole: Microsoft. The workaround is straightforward on its surface. Microsoft has poured 13 billion dollars into OpenAI since 2019, securing exclusive cloud hosting rights and enterprise distribution through Azure. That arrangement gave the tech giant its own licensing terms for OpenAI models, terms that didn't necessarily mirror OpenAI's consumer-facing restrictions.
This is more than a technical detail. The Defense Department allegedly experimented with Microsoft's Azure-hosted version of OpenAI technology while the ChatGPT maker's military prohibition remained in place. When OpenAI employees discovered this, it triggered uncomfortable questions about how corporate policy actually functions when advanced technology flows through multiple partners with divergent rules.
OpenAI's approach to military use shifted markedly over the course of 2024. OpenAI publicly maintained its military use ban until January 2024, when the company quietly updated its usage policies to allow defence applications. The change was not widely publicised. Instead of announcing a major policy reversal, OpenAI simply rewrote its usage terms, replacing explicit language about military and warfare with broader principles about avoiding harm. CEO Sam Altman later confirmed the shift, stating that OpenAI would work with the U.S. government on cybersecurity and other national security projects.
What makes the Wired report significant is not merely the policy change itself. It raises a structural problem in how technology governance works. When a company's AI is licensed and distributed by multiple partners, each operating under different terms, who actually controls the policy? It's about what happens when cutting-edge AI gets distributed through complex corporate partnerships.
This tension between stated policy and operational reality is worth examining carefully. From a fiscal and security standpoint, there is a legitimate case for Australia and its allies accessing the most advanced AI tools available. Competitive advantage in defence technology is real, and falling behind matters. But that argument only strengthens the need for transparency and clear policy frameworks. If the Pentagon was already testing OpenAI models through Azure, the public should have known. If companies intend to allow military use, they should say so plainly rather than relying on vague language or contractual technicalities.
The broader issue is not whether governments should access powerful AI tools. They should. The issue is how that access is governed and how accountability is maintained. The story's power comes from what it reveals about AI policy enforcement in an era of complex corporate partnerships and cloud-based distribution. What does it mean to ban military use when your technology is embedded in enterprise platforms with millions of users and hundreds of use cases? How do you enforce restrictions when third-party distributors operate under different rules?
For Australia, which increasingly relies on partnerships with these American tech giants, the lesson is clear. Ensure that any defence collaboration with AI companies is conducted with explicit terms, full transparency, and regular public accounting. Avoid relying on corporate policies that can shift quietly or be circumvented through subsidiary arrangements. Democratic nations deploying powerful technology deserve governance structures that are as sophisticated as the technology itself.