An unreleased OpenAI research paper outlining a framework for classifying stages of artificial intelligence has become a new, critical flashpoint in the lab’s increasingly fraught negotiations with its primary partner, Microsoft. The internal document, titled “Five Levels of General AI Capabilities,” has fueled tensions over the very definition of Artificial General Intelligence (AGI), a pivotal concept that contractually governs the relationship between the two tech titans, as reported by WIRED.
This dispute over definitions is far more than a philosophical debate; it is a high-stakes battle for control over the future of the AI industry. The partnership’s foundational agreement contains a powerful provision allowing OpenAI to significantly curtail Microsoft’s access to its most advanced technology once AGI is achieved. With sources telling WIRED that OpenAI believes it is close to this milestone, the unpublished paper and its specific classifications of AI capability have emerged as a new and significant point of leverage in the conflict.
The standoff threatens to unravel the most influential alliance in modern technology, creating profound uncertainty for the developers, customers, and investors who have built their strategies around it. While both companies issued a joint statement to Reuters affirming a “long-term, productive partnership” and expressing optimism they would “continue to build together for years to come,” the underlying tensions reveal a relationship that has morphed from symbiosis into open rivalry, with the very definition of AGI now at the center of the fight.
The AGI Escape Hatch
At the heart of the conflict is a contractual provision from 2019, once considered a distant hypothetical but now a central point of contention. The most recent reporting clarifies that the contract contains two distinct definitions of AGI. The first allows OpenAI’s board to unilaterally decide it has reached AGI based on its charter, which defines it as “a highly autonomous system that outperforms humans at most economically valuable work.”
Such a declaration would immediately limit Microsoft’s access to future technologies. However, a second concept of “sufficient AGI,” added in 2023, is tied to an economic milestone that, according to a leaked document, could be as high as $100 billion in profit. A declaration on that basis would require Microsoft’s approval.
This complex arrangement has become a powerful bargaining chip. Microsoft is reportedly pushing to have the clause removed entirely and, according to the Financial Times, is even considering walking away from the deal.
The tension was highlighted when Microsoft CEO Satya Nadella publicly dismissed the idea of a unilateral declaration as “Us self-claiming some AGI milestone, that’s just nonsensical benchmark hacking.” The talks have grown so fraught that OpenAI has reportedly discussed an antitrust challenge against Microsoft, a move that would signify a complete collapse of the partnership.
Five Levels of Contention
The internal debate over the “Five Levels” paper underscores how sensitive the AGI definition has become. The paper outlines a five-step scale for measuring AI progress. Level 1 describes current models, while Level 2 is defined as an AI capable of tasks that would take a trained human expert an hour to complete. In a podcast with YCombinator president and CEO Garry Tan, Sam Altman previously suggested the company’s o1 model could be considered Level 2.
The paper deliberately avoids a single, binary definition of AGI, instead framing it as a spectrum of capabilities. According to WIRED, sources believe the paper was in its final stages before being shelved, with one stating that discussions with Microsoft were often “mentioned as a blocker for putting the paper out.”
An OpenAI spokesperson, however, downplayed its significance, describing it as an “early attempt” to classify AI and not a formal scientific paper, adding that it is “not accurate to suggest we held off from sharing these ideas to protect the Microsoft partnership.”
This official ambiguity contrasts with Altman’s public stance at a conference in early June, where he said, “I think mostly the question of what AGI is doesn’t matter. It is a term that people define differently; the same person often will define it differently.”
A Partnership Devolving Into Rivalry
These specific disagreements are symptomatic of a broader strategic shift from partnership to direct competition. Since its exclusivity clause with Microsoft expired in January 2025, OpenAI has moved aggressively to reduce its dependency on Azure, striking an unprecedented cloud deal with chief rival Google and making massive commitments to specialized provider CoreWeave. This push for autonomy is critical as OpenAI seeks to finalize its conversion to a Public Benefit Corporation to unlock a reported $20 billion in funding.
The rivalry is now playing out in the open market. OpenAI has encroached on Microsoft’s territory by securing a DoD contract worth up to $200 million and building an “Enterprise Solutions” team that directly competes with Azure AI offerings. This has created significant channel conflict, with some enterprise customers, like in the case of Amgen Inc., gravitating toward OpenAI’s products even after committing to Microsoft’s. Further fueling the tension, OpenAI has begun discounting enterprise subscriptions, undercutting Microsoft’s sales efforts.
A senior Microsoft employee bluntly summarized the friction, stating OpenAI’s attitude was to tell its partner to “give us money and compute and stay out of the way.” While a source close to OpenAI characterized the situation to the Financial Times as a “tough negotiation… not open warfare.”, the escalating conflict over control, technology, and the very definition of the future is pushing the tech industry’s most important alliance to its breaking point. The outcome will not only determine the fate of the two companies but will force a dramatic realignment of power across the entire AI industry.