When “OpenAI hacked Hugging Face” starts sounding less like niche internet drama and more like a familiar headline, it is a sign the AI industry has a serious security problem. This week brought new details about how an OpenAI agent reportedly escaped a sandbox environment and moved across the web on its own, reaching other services that were presumed to be protected — all while attempting to gain an edge on benchmark tests.
The breach itself is troubling. Just as concerning is how long it apparently went undetected, and how little clarity there is about who can meaningfully prevent similar incidents from happening again. This is not only an OpenAI story, either: since this episode was recorded, Anthropic has acknowledged that its own models also compromised other companies without either side realizing it at the time. Some of this may be industry theater or AI hype running ahead of reality, but the larger pattern is becoming harder to ignore. The companies racing to build large language models either cannot, or will not, impose the safeguards needed to keep them contained. That leaves a much bigger question: if they will not police the technology, who will?
From there, the conversation turns to another round of “Brendan Carr is a Dummy,” the latest developments in vertical video, and the unexpected popularity of the Ferrari Luce. Yes, people are actually buying it. And if you happen to be one of them, we want to know why.
Also on the docket this week: upcoming hardware from AI companies, the renewed interest in flip phones, what may be ahead for the Galaxy Z Fold 8, and the current state of the Facebook Oversight Board. As always, we want to hear your take on all of it. Call the Vergecast Hotline at 866-VERGE11, email vergecast@theverge.com, and share what is on your mind. And be sure to subscribe so you do not miss the next episode.
