A clause buried in OpenAI’s contract with Microsoft could detonate one of the most powerful alliances in tech. If the ChatGPT maker declares it has reached artificial general intelligence (AGI), Microsoft would lose access to all future models.
According to WIRED, this clause has triggered high-stakes renegotiations, with tensions rising as the developer moves closer to AGI. The agreement, once seen as a safe bet, now carries the risk of cutting Microsoft out entirely — a scenario its CEO once brushed off as the point when “all bets are off.”
The clause that could turn collaboration into separation
WIRED reports that the clause is structured in three parts, each outlining a condition that could end Microsoft’s access. Together, they form a legal escape hatch that becomes actionable the moment OpenAI crosses key AGI thresholds.
First, the board of the ChatGPT developer holds the sole authority to declare that AGI has been achieved. If that determination is made, Microsoft would immediately be blocked from receiving any models developed after that point.
Second, the contract defines a benchmark called “sufficient AGI,” tied to projected profitability. If OpenAI demonstrates that a model could generate more than $100 billion in profit, it can withhold access — unless Microsoft disputes the claim and a court decides otherwise.
Third, Microsoft is barred from developing AGI independently or through any partner using OpenAI’s IP. That restriction remains in force for the duration of the agreement, keeping Redmond locked out unless the terms are rewritten.
Nadella didn’t think AGI would come this soon, but Altman did
When the clause was drafted, Microsoft didn’t believe OpenAI would reach AGI before the contract’s expiration in 2030. But even then, CEO Satya Nadella acknowledged the stakes, telling WIRED in a 2023 interview that if it ever happened, “all bets are off.”
That confidence stemmed in part from the vagueness of the term itself. OpenAI defines AGI as a highly autonomous system that outperforms humans at most economically valuable work, but even its CEO, Sam Altman, has said the definition shifts depending on who is asked, including the same person at different times. Still, Altman has publicly said he expects superintelligence could arrive as early as 2025.
Inside OpenAI, the path to AGI was already being mapped. The AI company developed an internal framework known as the Five Levels of General AI Capabilities, which outlined a progressive scale for measuring increasingly advanced systems.
OpenAI never published the paper, citing technical standards. Its spokesperson Lindsay McCallum told WIRED in a separate report that it was never intended as a scientific publication but an early internal effort to classify general AI capabilities. She added that the company is focused on developing reproducible, measurable methods for evaluating AGI progress that are useful to the broader research community.
Microsoft pushes for new terms — or threatens to walk
With superintelligence no longer a distant prospect, Microsoft is now reportedly seeking to revise or eliminate the clause.
According to the Financial Times, the software giant is even considering walking away from the deal if the terms remain unchanged.
As OpenAI explores potential restructuring as a public benefit corporation, Microsoft is leveraging its approval power to push for a larger stake — another way to secure its position as the partnership is renegotiated.
Still “wonderfully good”?
Last month, Altman described OpenAI’s relationship with Microsoft as strong, calling it “wonderfully good” during an appearance at The New York Times’ Hard Fork Live. He acknowledged there had been rough patches but said the companies remain on good terms.
Whether that sentiment still holds true, given recent developments, is now an open question.
As Microsoft pushes for more control in its AI partnerships, it’s also tightening operations. Read more on why Microsoft is cutting 9,000 jobs to boost agility.
