OpenAI’s head of safety, Joshua Achiam, is out.
After nearly nine years digging into AI risk research—and making quite the impression during the Musk versus Altman trial—Achiam is packing his boxes. It’s a clean break, mostly. The futurist chief title is also departing.
Across the Bay Area, things are uglier.
Meta employees are fed up with Mark Zuckerberg’s idea of a company-wide AI hackathon. One staff member posted openly for everyone to see: “I’m not sure that this company supports hackathon culture anymore.” That’s not just a complaint. It’s an indictment of a workforce burned by endless reorgs.
The chaos is spilling over into privacy too.
Meta paused an internal employee-tracking program after sensitive data leaked inside the company. Oops. They left the info exposed. Then they had to scrub it. It doesn’t look good when the surveillance camera turns on itself.
“I’m not sure that this company has the appetite for this right now,” isn’t what anyone is saying, but the silence is loud enough.
At OpenAI, the gate is slamming shut.
They’ve built new AI models—GPT-5.6 sounds flashy on paper. You can’t touch them, though. The White House intervened, asking for a delay on the rollout. This comes just two weeks after Anthropic had to yank its most advanced models offline. The government isn’t playing around anymore. Or maybe the labs are finally scared.
Back at Meta, the pain is visceral.
CTO Andrew Bosworth wrote an internal memo admitting their recent AI restructuring was “atrocious.” Strong word for a tech exec. He’s promising stability. Better comms. The return of perks. He knows morale is in the toilet.
Stability, he says, is coming.
We’ll believe it when we see the org chart stop shifting every Tuesday.
Google DeepMind is trying something harder.
Unionization talks are stumbling out of the gate. Employees say executives aren’t just reluctant to unionize—they’re unwilling to even talk about it meaningfully. Frustration is brewing in the negotiation room.
Meanwhile, inside OpenAI, someone has to build the product while the lawyers panic.
Meet Thibault Sottiaux. He helped make AI coding into OpenAI’s growth engine. Now he’s shepherding ChatGPT through its biggest transformation yet. It’s messy engineering. It’s vital work. It’s keeping the lights on.
Is Cursor next?
SpaceX bought Cursor, the code editor. There’s a real question of whether SpaceX will let Cursor keep hosting models from rivals like OpenAI and Anthopic. It tests the fragile friendships between the big AI labs. Can an Elon company hold neutral ground?
Doubtful.
At Meta, the AI unit is a train wreck.
Leaked discussions call it “a total mess.” One anonymous source dropped a gem: “Tell him he’s a piece of shit.” That’s corporate governance gone wrong. Executives are lost. Employees are screaming. The strategy has no shape.
On the hardware front, big money changes hands.
Qualcomm bought Modular, the buzzy chip software startup, for nearly $4 billion. One of the hottest bets of the AI hardware era just cashed out.
And at the edge, users are noticing the bill.
Token usage is skyrocketing. “Tokenomics” isn’t just a buzzword; it’s a budget crisis. Software makers and e-commerce brands are struggling to price this thing out. Who pays when the context window gets too heavy?
No one really knows yet.
