Behind OpenAI’s Takeover of Thinking Machines Lab

If a series from HBO Max ever emerges focusing on the AI sector, this week’s events will certainly make for an intriguing episode.

On Wednesday, OpenAI’s application CEO, Fidji Simo, revealed that the company has rehired Barret Zoph and Luke Metz, the cofounders of Mira Murati’s AI venture, Thinking Machines Lab. Zoph and Metz had departed OpenAI in late 2024.

We shared insights last night regarding two narratives that surfaced about the reasons behind their departures, and we have now obtained further details.

A source with firsthand knowledge stated that the leadership at Thinking Machines believed Zoph was involved in a serious misconduct incident while at the company last year. This event reportedly eroded Murati’s trust, according to the source, and disturbed their working dynamic. Additionally, the source claimed that Murati terminated Zoph on Wednesday—before becoming aware of his return to OpenAI—due to issues that arose following the alleged misconduct. Around the time the company discovered Zoph’s return to OpenAI, internal concerns were raised about whether he shared confidential information with competitors. (Zoph has not replied to multiple requests for comment from WIRED.)

In a memo to employees on Wednesday, Simo stated that the rehiring plan had been in the works for weeks and that Zoph informed Murati of his intention to leave Thinking Machines on Monday—prior to his termination. Simo also assured employees that OpenAI does not share Thinking Machines’ concerns regarding Zoph’s ethics.

Alongside Zoph and Metz, another former OpenAI researcher at Thinking Machines, Sam Schoenholz, is also rejoining the ChatGPT creator, as per Simo’s announcement. Two additional employees from Thinking Machines are expected to join OpenAI in the upcoming weeks, according to a source with knowledge of the situation. Technology reporter Alex Heath first reported on these potential hires.

Another source familiar with the situation countered the notion that the recent personnel shifts were solely related to Zoph. “This has been part of a long conversation at Thinking Machines. There were discussions and misalignments regarding the company’s vision—it involved the product, the technology, and the future.”

Neither Thinking Machines Lab nor OpenAI chose to comment.

Following these developments, multiple researchers at top AI labs have expressed their exhaustion with the ongoing turmoil within their industry. This particular incident echoes OpenAI’s brief removal of Sam Altman in 2023, a situation known internally as “the blip.” Murati played a significant role in that event as the company’s then chief technology officer, according to reports from The Wall Street Journal.

Since Altman’s departure, the turbulence in the AI sector has persisted, including the exits of cofounders from various major AI labs, such as xAI’s Igor Babuschkin, Safe Superintelligence’s Daniel Gross, and Meta’s Yann LeCun (who co-founded Facebook’s enduring AI lab, FAIR).

Some may argue that the drama is warranted for a burgeoning industry whose investments are contributing to the growth of America’s GDP. Furthermore, if you subscribe to the notion that one of these researchers may unlock breakthroughs on the path to AGI, tracking their movements seems beneficial.

Nevertheless, many researchers, who began their work prior to ChatGPT’s explosion in popularity, appear surprised by the relentless scrutiny now faced by their industry.

As long as researchers can continue to attract billion-dollar seed rounds at will, it seems likely that upheavals in the AI sector will persist. HBO Max writers, get ready.

How AI Labs Are Training Agents to Do Your Job

In Silicon Valley, discussions about AI replacing jobs have been ongoing for decades. However, recently, the efforts to enable AI to perform commercially valuable tasks have become significantly more advanced.

AI labs are becoming more discerning about the data used to develop AI agents. Last week, WIRED reported that OpenAI has asked third-party contractors from the firm Handshake to upload examples of their actual work from previous positions to assess OpenAI’s agents. Companies require employees to remove any confidential information and personally identifiable details from these documents. While it’s possible that some corporate secrets or names could inadvertently slip through, that’s likely not the main intention of OpenAI (although experts warn that serious consequences could arise if that were to happen).