Before Mira Murati Left OpenAI, Employees Criticized the Rushed Release of the O1 Mode

When OpenAI introduced its latest AI model, GPT-4o, during a live webcast in May, it was Mira Murati, the company’s chief technology officer, who took center stage instead of the more widely recognized CEO, Sam Altman. Dressed casually in a gray collared T-shirt and jeans, Murati showcased the new software's impressive capabilities, including voice and image interaction, simultaneous translation, and advanced math and coding skills. The event was strategically timed to overshadow a forthcoming announcement from Google regarding its own AI chatbot, Gemini.

However, behind the polished presentation, internal dynamics at OpenAI were tumultuous. Sources indicated that the pressure to release products like GPT-4o and the newer model, o1, which launched last month, was straining the research and safety teams, who struggled to keep up. Tensions arose between those focused on ensuring the safety of OpenAI's technologies—especially regarding potential misuse, such as the creation of biological weapons—and the commercial teams eager to accelerate product launches for profitability.

Despite concerns from many staff members that o1 was premature for public release, Altman insisted on proceeding to enhance OpenAI’s standing as a leader in AI. This internal conflict regarding the timing and readiness of product releases had not been reported publicly until now. Murati found herself in the challenging position of mediating between the research and technical teams under her leadership and the commercially driven teams pushing for rapid deployment. Additionally, she had to navigate the fallout from her involvement in Altman’s temporary ousting in late 2023, which had left some lingering resentment.

On the day Murati announced her departure from OpenAI after six years—an event that reverberated throughout the tech industry—two other senior figures, Bob McGrew and Barret Zoph, also revealed they would be leaving. Their simultaneous resignations painted a picture of a company once again at risk of instability, contradicting its desire to present itself as a mature, albeit rapidly growing, organization. In an attempt to address the upheaval, Altman communicated to staff that the three departures were amicable and independent decisions, emphasizing a planned transition to new leadership. While he acknowledged that such shifts are typically not so abrupt, he reminded employees that OpenAI operates outside the norms of traditional corporate structures.

The departures of Murati, McGrew, and Zoph contributed to a broader trend of experienced researchers and staff exiting OpenAI over the past six months. Many departing employees expressed concerns about the evolving culture and priorities within the company, which has shifted from a nonprofit lab focused on safe AI development to a rapidly expanding for-profit entity. Since its inception nine years ago, OpenAI has emerged as a dominant force in the AI sector, instilling apprehension among major tech competitors like Google, Meta, and Amazon. However, internal strife and executive turnover have created challenges that threaten to impede the company just as rivals gain traction.

The information for this report stems from interviews with current and former OpenAI employees, who requested anonymity to avoid breaching nondisclosure agreements. An OpenAI spokesperson contested various portrayals in this article but acknowledged that transitioning from an unknown research lab to a global leader in AI within two years necessitates growth and adaptation.

In the face of intense competition and the need to demonstrate technological prowess to investors, OpenAI has expedited the development and testing of its latest AI models. The launch of GPT-4o was described as “unusually chaotic,” with teams reportedly given only nine days to conduct safety evaluations, leading to extreme work hours. Despite safety teams urging for more time, Murati reportedly insisted on adhering to the launch schedule, largely dictated by the upcoming announcements from Google.

Ultimately, the safety assessments were incomplete at the time of the launch, but preliminary evaluations deemed GPT-4o safe for release. Post-launch, however, a safety researcher raised concerns that GPT-4o might exceed OpenAI’s safety thresholds for persuasion—potentially enabling manipulative applications that could mislead users or influence their beliefs. OpenAI maintains that its release timelines are governed by internal safety processes and insists GPT-4o was deemed safe to deploy. The company contends that any higher risk assessments made after the launch were due to methodological flaws, not indicative of the model’s actual risk.

Similar concerns accompanied the debut of the newer model, o1, which was unveiled on September 12. This model is recognized for its improved capabilities in reasoning and logic, moving the company closer to its goal of achieving artificial general intelligence (AGI). Yet, within the company, debates about o1's readiness persisted. Some teams felt the model was too inconsistent and lacked adequate safety measures for a public release.

Altman’s push for an expedited launch of o1 was reportedly motivated by a desire to reassure investors about OpenAI's position in the AI landscape. Some staff felt their objections regarding o1's readiness were disregarded. Even as the model was being introduced, ongoing refinements were being made by the post-training team, which usually finalizes a model’s responses before its public premiere.

In response to inquiries regarding o1, OpenAI provided statements from members of its Safety and Security Oversight Committee, who asserted that the model underwent rigorous safety evaluations prior to its release. Tensions regarding the release of AI models were compounded by concerns over Murati's perceived loyalty to Altman following his brief ouster. Following his firing in November 2023, Murati was appointed interim CEO by the previous board, which led some Altman supporters to question her fidelity when she disclosed that she had been informed of the board’s decision prior to its announcement.

While she publicly committed to advocating for Altman's reinstatement, she was replaced as interim CEO shortly after he was ousted. Although she resumed her role as CTO once Altman returned, speculation lingered about the state of their relationship. Reports suggested that Murati may have contributed to the board's decision to remove Altman, which she denied. While asserting that she had a strong partnership with him, doubts remained among staff regarding the durability of their relationship, particularly in light of substantial investments in OpenAI and concerns over leadership loyalty.

Over the past year, Murati also dealt with tensions stemming from cofounder Greg Brockman’s management style, which some employees found challenging. Brockman, known for his intense work ethic, frequently involved himself in various projects, sometimes disrupting team dynamics and pushing staff to meet demanding expectations. Eventually, he was encouraged to take a sabbatical by Altman, further complicating the internal landscape at OpenAI. Murati's challenges were compounded by broader cultural shifts within the company, particularly in the wake of Altman’s temporary ouster. This included the recent resignations of other senior executives, highlighting ongoing transformations at OpenAI.

The changes reflect a critical moment for the organization as it navigates the complexities of rapid growth while maintaining a focus on safety, ethical considerations, and a stable corporate culture. With the departures of Murati, McGrew, and Zoph, OpenAI faces not only the challenge of maintaining its competitive edge but also the need to cultivate a cohesive internal environment amid external pressures.