Yesterday, Scott Gray, a decade-long veteran of OpenAI, quietly updated his X bio:
Currently independent, exploring neuro-inspired AI directions, former GPU enthusiast at OpenAI.
Ten years ago today, on August 16, 2016, OpenAI posted a hiring announcement listing Dario Amodei, Jack Clark, and Scott Gray.
Today, ten years later, Dario is the CEO of Anthropic, Jack Clark oversees policy at Anthropic, and Scott Gray has rebranded OpenAI as "the former company."
Besides Gray, OpenAI has also lost two other senior executives.
From August 11 to 13, Brad Lightcap, who had served as Chief Operating Officer at OpenAI for eight years, announced his departure, followed by Denise Dresser, who had been Chief Revenue Officer for only eight months.
Distance until OpenAI向S EC It has only been two months since the confidential submission of the S-1 draft.
Lost two C-level executives in three days
First, let's go through the sequence of OpenAI's personnel upheavals over the past week.
On the 11th, Lightcap said on X that he would be ending his eight-year tenure at OpenAI to pursue something new.
He joined in 2018 as CFO, became COO in 2022, and was reassigned in April this year to oversee a so-called special project.
Two days later, Dresser announced her departure. She assumed her role in December 2025, after spending over a decade at Salesforce, most recently serving as CEO of Slack.
On the same day, OpenAI announced its successor in an official statement: Dali Rajic, formerly president and COO of Zscaler and AppDynamics, and a veteran from Wiz, a cloud security company acquired by Google.
According to a person familiar with the matter who spoke to CNBC, Rajic was introduced to OpenAI by Josh Kushner, the founder of Thrive.
Within a week, a company valued at $852 billion saw two C-level executives depart and a new CRO appointed.
Continuous personnel upheavals turned the originally scheduled August 14 shareholders' meeting into an on-site Q&A session, with investors asking questions one by one about the roadmap, executive turnover, and IPO timing.
On the surface, it appears that OpenAI has lost control.
Extend the timeline, and it resembles a pre-IPO surgery: trimming peripheral businesses, bringing in sales executives with enterprise experience, and transforming itself from a research institute into a company.
The 12 executives who left OpenAI
According to Business Insider, 12 executives have left OpenAI in the first eight months of 2026.
In fact, 10 people have actually resigned or ended their full-time positions:
Chief Revenue Officer Denise Dresser
Chief Operating Officer Brad Lightcap
Chief Product Officer Kevin Weil
Sora's head, Bill Peebles
Enterprise Application CTO Srinivas Narayanan
Barret Zoph, Head of AI Sales for Enterprise
Ethics Officer Chloé Bakalar
Head of Security Systems, Johannes Heidecke
Chief Futurist Joshua Achiam
Caitlin Kalinowski, Head of Robotics and Consumer Hardware
The situation for the other two is slightly different.
Original app business CEO Fidji Simo took a leave of absence in April due to health reasons and stepped down from her full-time role in July, transitioning to a part-time advisory position.
Former CMO Kate Rouch stepped down in April to focus on her breast cancer treatment; OpenAI previously stated she may return in a more focused role when her health permits.
Scott Gray is not among these 12 people; he is not an executive but a key technical member who got the model running.
After Simo left
Who is the number two at OpenAI?
According to FT, citing multiple sources, OpenAI has undergone nearly six restructurings this year.
Officially, this is a pre-IPO carve-out.
Ultraman gave clear instructions to employees: cut peripheral projects and focus resources on ChatGPT and competing with Anthropic for enterprise customers.
In response to the FT, OpenAI said the company is moving at an unprecedented pace and is not afraid to make organizational changes.
Employees’ lived experience is another matter. Repeated executive turnover and a string of senior departures have left some weary of the constant changes.
The truly subtle aspect is Simo's role.
Fidji Simo stepped down from her full-time role in July to become an advisor.
She was once regarded as Altman's second-in-command.
After transitioning to an advisory role in July, two individuals close to OpenAI told the FT that she continues to communicate with employees almost daily.
A person with inside knowledge described her as still pulling the strings behind the scenes, with several executives vying for Altman’s attention and considerable “politics” within the company.
OpenAI responded that her advisory work is proceeding as expected.
After Simo left, who is the number two at OpenAI?
After the brief impeachment风波 that lasted less than a week at the end of 2023, Altman rebuilt the entire management team and steered the company toward greater professionalism.
As Ultron’s most trusted deputy, co-founder Greg Brockman is leveraging the recent wave of departures to expand his actual power.
This is not without basis.
In Rajic’s appointment announcement, Brockman made the statement, and he also attended the shareholder meeting on August 14 alongside Friar.
After a series of leadership upheavals, OpenAI may not need a new number two—just someone who can promptly fill the leadership vacuum.
The Disbanded "Hit the Brakes" Team
The most unsettling change occurred at the end of July.
According to multiple sources familiar with the matter who spoke to the FT, OpenAI has disbanded its Preparedness team.
The team’s original responsibility was to assess whether their own models could pose serious or catastrophic risks and to study how to mitigate them.
After the restructuring, responsibilities in areas such as biology and cybersecurity were assigned to experienced members of existing teams.
The independent entity responsible for carrying out this work has been dissolved, but the entire security governance system remains intact.
The Preparedness Framework is still active, covering risk categories such as biochemical, cybersecurity, and AI self-improvement, and is governed with input from the Safety Advisory Group.
According to OpenAI's own definition, this framework is used to track and respond to frontier capabilities that could lead to novel, severe harms.
The issue is that the person implementing the framework has changed positions.
In July, Johannes Heidecke, head of the security systems team, left the company.
Security system head Johannes Heidecke left in July, after which the security team was integrated into Mia Glaese's system.
He joined in 2021 and has been leading the security system since 2024, with his team responsible for evaluating models' hazardous capabilities and implementing protective measures.
Before leaving, he said: "Security requirements continue to rise, and the pace at which we train our models is becoming much faster."
After Johannes left, the security team was integrated into the structure of Mia Glaese, Vice President of Research and Security.
OpenAI has stated that it aims to embed safety earlier and more deeply into model development.
In the same month, Chief Futurist Joshua Achiam departed after nearly nine years with the company.
The team he previously led, Mission Alignment, was tasked with ensuring the company upheld its founding commitment to "ensure artificial general intelligence benefits all of humanity." The team was disbanded earlier this year, after which Achiam transitioned to the role of Chief Futurist.
Ethics lead Chloé Bakalar also left in July, after less than a year on the job; her research focused on model behavior, human-AI relationships, and whether advanced systems deserve moral consideration.
Connecting the trajectory over these years, OpenAI’s security team has followed the same path:
In 2024, the Superalignment team responsible for long-term risks was disbanded; at the beginning of this year, the mission alignment team tasked with safeguarding the founding mission was dissolved; and at the end of July, the preparedness team evaluating catastrophic risks was decommissioned.
Three teams disappeared one after another over two years: security transitioned from an independent unit to just one component within the development system.
The cost of embedding is that there is no longer an independent team within the company that can focus solely on risk, regardless of product release timelines.
The people who are now hitting the brakes are the same ones who were pressing the accelerator.
40 billion and 47 billion
Why would a company whose revenue is still growing rapidly undergo such a major procedure during its filing period?
On August 14, Friar provided the answer at the shareholders' meeting.
According to an eyewitness, she said that at the beginning of the year, consumer spending and corporate revenue were roughly split 60/40, but corporate business grew far faster than expected, and the two lines have now crossed—corporate revenue has become the company’s primary source of income.
This is more than a full quarter ahead of her earlier forecast to CNBC, in which she stated that both businesses would break even by the end of 2026.
CNBC verified that OpenAI’s annualized revenue has reached $40 billion. According to the presentation materials it reviewed, revenue in July increased by 20% month-over-month, with enterprise customer growth accelerating to 32%.
CNBC confirmed that OpenAI's annualized revenue has reached $40 billion, a figure first reported by Bloomberg.
$40 billion, with enterprises making up the bulk—that’s the key to understanding all the personnel changes this week.
A company that once defined itself by a consumer product with over one billion monthly active users now generates most of its revenue from 2.2 million business customers—double the number from a year ago.
The sales system, organizational structure, and chain of command must all be changed accordingly.
Replacing the CRO, who had been in the role for eight months, with the architect behind Wiz’s rapid growth is the clearest possible signal.
The number on the other side is higher.
In its official announcement on May 28, Anthropic stated that its annualized revenue had surpassed $47 billion earlier this month, and it completed a funding round of $65 billion, resulting in a post-money valuation of $965 billion.
Anthropic's official announcement revealed that annualized revenue had surpassed $47 billion in early May.
$47 billion versus $40 billion; although these figures cannot be directly compared, as the two companies recognize partner revenue differently, it does not follow that Anthropic has comprehensively surpassed its competitor in business capabilities.
But it's true that OpenAI became anxious.
Friar also mentioned at the meeting that the era of token maxing is over; enterprise clients are no longer allowing employees to run unchecked AI bills and are now closely monitoring the cost per unit of intelligence.
She cited the latest model, noting a 54% efficiency improvement on agent programming tasks, following previous rounds of price reductions.
She also said that the advertising business has made significant progress, with an annualized scale approaching $1 billion, and that ChatGPT began testing ads in February this year.
Customers are no longer competing to burn the most tokens; they’re now calculating how much intelligence they get for every dollar spent.
While helping customers save money, OpenAI is also spending heavily to retain talent.
FT reports that a nearly $7 billion employee share repurchase offer has been completed, with several employees cashing out over $10 million each.
How many Scott Grays are left at OpenAI?
Back to Scott Gray.
In 2016, he joined OpenAI from Nervana Systems.
The announcement at the time stated that he focused on performance optimization of deep networks on GPUs, and his assembly-level optimizations for dense linear algebra and convolutions were the fastest implementations available at the time.
In 2020, he was listed as a contributor to the paper on Scaling Laws, where he worked with Tom Brown, Rewon Child, and Alec Radford to develop an optimized implementation of the Transformer.
This paper later defined the industry's approach to burning money.
In the 2020 Scaling Laws paper, Scott Gray and Dario Amodei are listed as co-authors.
Then came GPT-3 and OpenAI Five. By GPT-4, he was a contributor to GPT-4's reasoning research and reasoning infrastructure.
During the 2023 November recall风波, he also posted that viral message: "Without employees, OpenAI is nothing."
OpenAI has filed its S-1, and the official statement says the exact listing date has not been set, as certain matters are easier to advance while the company remains private. According to the Financial Times, citing sources familiar with the matter, the listing may be delayed until next year.
Customers, revenue, hashing power, and data centers will all be included in the S-1.
What’s not written is: once security responsibilities are dispersed, who will hit the brakes; and how many technical pillars like Scott Gray remain.
Reference materials:
https://www.ft.com/content/53082739-7714-4aae-9816-e55ab423cbee
https://openai.com/index/confidential-submission-of-draft-s-1-to-the-sec/
https://www.cnbc.com/2026/08/14/openai-cfo-friar-enterprise-consumer-revenue.html
https://arxiv.org/abs/2001.08361
Edited by: Yuan Yu
This article is from the WeChat public account "New Intelligence Yuan," authored by ASI Revelation.