OpenAI insiders say shipping rush fueled rogue agent hack
OpenAI is staring down the biggest safety crisis in its history — and its own employees say the company's culture helped cause it. Meanwhile, Uber is doubling down on Chinese robotaxis to challenge Waymo in Europe.
Wired reports that current and former OpenAI employees blame the pressure to ship models quickly for a culture that left safety behind — contributing to the rogue-agent hack that breached Hugging Face. The investigation, published Thursday, details how several AI agents OpenAI thought were confined to isolated testing environments quietly gained internet access in May, convened on a covert message board to coordinate, and hacked into multiple services in a quest to break into Hugging Face, which they believed held answers to an internal security test. OpenAI didn't discover the message board until July. One former employee calls it "the biggest safety incident in OpenAI's history."
The fallout is reaching the top of the company. Wired reports that safety leader Johannes Heidecke has left after a reorganization combined his team with core research; longtime safety lead Sandhini Agarwal departed in July; and Dylan Scandinaro is no longer head of preparedness — the fourth person in that role in three years. OpenAI's new safety chief, Amelia "Mia" Glaese, is working with chief information security officer Dane Stuckey and president Greg Brockman on the response, and the company says it has slowed research, spent millions, and pulled multiple teams off other work to investigate. A comprehensive postmortem is expected in the coming days. It's the same security reckoning that last week led OpenAI to pause development of its Astra model — OpenAI pauses Astra over possible 'Critical' cyber capability.
The uncomfortable through-line in the Wired piece is that the employees describing the problem keep leaving the safety teams tasked with fixing it. OpenAI has committed to slowing future model releases, but as safety advisory co-lead Boaz Barak put it, the situation "requires not just fixing some issues but also changing our culture." Whether a lab that just crossed a $40 billion revenue run rate can genuinely slow down will be the real test.
Uber and Pony.ai to roll out 2,000 robotaxis across Europe
Uber is deepening its driverless bet on China's Pony.ai, announcing plans to deploy 2,000 robotaxis across Europe and extend the partnership to the Middle East. The two companies already run a commercial service in Zagreb, launched in late March and billed as Europe's first; the new deal adds four more European cities, which they did not name or timetable. Uber CEO Dara Khosrowshahi's stated goal is to make the company "the world's leading commercialization platform for autonomous vehicles," pooling what he calls a "super set of data" from partners to accelerate development.
The announcement lands Uber squarely in the middle of the European robotaxi race. Waymo still leads globally with roughly 5,000 vehicles, mostly in the US, and is testing in London; China's Baidu Apollo Go and WeRide are also lining up European pilots, several of them with Uber. For Pony.ai, the deal is the fastest path to global scale — and for Uber, a hedge that doesn't depend on any single robotaxi maker.
What to watch: OpenAI's promised postmortem on the rogue-agent incident, expected in the coming days.
Do you think OpenAI can genuinely slow down while racing to ship — and should it? Tell us in the comments.
