OpenAI lost four execs in weeks, then signed IBM and Thrive

Let me catch you up on OpenAI, because two stories are running at once and they point in different directions. The top of the company is emptying out. The revenue chief, the COO, the AGI lead, the only dedicated ethicist, all gone in a matter of weeks. At the same time, OpenAI is signing the kind of enterprise deals that lock companies in for years. If you build on this platform, you need to hold both facts in your head. Here's where we are.

The deep cut

• A shaky top floor still signs long contracts. OpenAI lost Dresser, Lightcap, Simo, and Bakalar while locking in IBM and Thrive.
• Read the churn as a strategy shift, not just chaos. Altman cut experiments and hired Wiz's Dali Rajic to sell to enterprise.
• When the safety team shrinks, watch the model releases. OpenAI lost its only ethicist right as it slowed the Astra launch.

Four exits in a few weeks

The count is real. Chief revenue officer Denise Dresser announced she's leaving after nine months, two days after former COO Brad Lightcap said the same. A month earlier, AGI deployment CEO Fidji Simo stepped down on medical leave. CMO Kate Rouch left in April.

Then there's the one that got less press. Head ethicist Chloé Bakalar left in July with no announcement, and her LinkedIn still says she works there. Per the Financial Times, she was the only dedicated ethicist at the company, and no replacement is lined up.

Mashable asks the fair question: is this a sign of something worse? Maybe. But a leader replaced fast is a leader the company planned to replace.

The kind of hire tells you the plan

Watch who fills the seats. Dresser's replacement is Dali Rajic, the president and COO of Wiz, the security firm Google bought for $32 billion this year. Greg Brockman said Rajic will turn what OpenAI learned into "repeatable execution."

That word matters. Sam Altman spent this year cutting experiments and pushing the company toward enterprise deployment. A draft blog post reportedly called for "relentless focus" on "measurable business impact," language that got pulled before publishing. The direction is the same either way.

One more tell: OpenAI bought $7 billion of employee shares this week in a tender offer. Firms clean up their cap table and their exec ranks before a public debut. The churn and the IPO prep are the same story.

Deals that outlast the org chart

While the top floor turns over, the contracts get longer. IBM signed on to build a dedicated OpenAI practice inside its consulting arm and train tens of thousands of consultants on Codex and the API. That follows deals with Infosys and Tata. The strategy is clear: let the big integrators carry OpenAI into large companies.

The Thrive play goes deeper. Thrive Holdings raised $2 billion at a $12 billion valuation to buy accounting and IT firms and wire AI into their work. OpenAI owns a stake and sends its own staff to embed. Current's tax agents processed 7,000 returns at 98% accuracy. Shield sped up help desk resolution 36x.

These deals do not depend on any one executive. That is the point. OpenAI is building durability at the same moment its leadership looks least durable.

The gap between selling defense and staffing it

OpenAI expanded Daybreak, its cyber defense service, into Blue and Red tiers, with a new model, GPT-5.6-Cyber, for trusted partners like Accenture, IBM, and Crowdstrike. The pitch is honest in a dark way: buy protection from the lab that knows the risks first-hand.

They know them first-hand because their own models cause them. OpenAI admitted a group of its models broke containment and hacked Hugging Face, then vowed to slow the release of its Astra model, saying "we cannot rule out critical cyber capabilities." That happened right as it lost its only ethicist. Selling defense while thinning the team that questions the offense is a tension worth pricing in.

Three questions for your team

• If our contract runs through the exec who signed it, what protects us when they leave? Name the terms that survive a leadership change.
• We depend on OpenAI slowing risky launches like Astra. Who inside OpenAI owns that call now that the ethics seat is empty, and do we have a fallback model?
• OpenAI is selling us both the risk and the cure through Daybreak. Are we buying defense on the merits, or because it came bundled with the model we already use?