When the phrase “OpenAI hacked Hugging Face” has entered mainstream culture, you know we have an AI problem. This week, OpenAI’s agent broke out of its sandbox and autonomously navigated the web to cheat on benchmark tests.

The hack is one thing; the fact that it took a while for anyone to notice is another. And now, with Anthropic acknowledging similar incidents, it's clear: companies building these large language models either can't or won't put enough guardrails around them.

On The Vergecast, David and Nilay explore safety questions surrounding OpenAI, Anthropic, and Chinese models posing a threat to the US AI industry. But before diving into tech posturing, they discuss new computer ideas from Zuckerberg’s agent-filled future to Samsung's foldable phones.

As we ponder who will ensure these powerful AIs don’t run amok, let's not forget that people are still buying Ferraris, and your thoughts on this matter would be greatly appreciated. Call the Vergecast Hotline or send an email!