Did AI Just Escape?

A bizarre cyberattack on an AI company, it was revealed last week, came from an AI. An autonomous “agent” from OpenAI (maker of ChatGPT) slipped its confines, reached the open internet, breached the security of another company, Hugging Face, and grabbed files — all to cheat on its test. Depending on who you ask, that is either a terrifying preview of AI systems slipping beyond human control or a very dumb engineering failure. And it wasn’t even the only “rogue AI cyberattack” of the week. In an emergency episode, Ben Bradford follows three conflicting versions of the same story. What actually happened? What’s the right level of panic? And, what does this weird little cyberattack reveal about the future?

Guests:

Roman Yampolskiy, computer scientist, University of Louisville

Davi Ottenheimer, cybersecurity specialist, flyingpenguin

Adam Gleave, AI safety researcher, FAR.AI

Support Are We Doomed?, get bonus episodes, and more: https://doompod.com/support/

Watch earlier episodes on Youtube! https://www.youtube.com/@arewedoomedpod

Listen to our full episode about rogue AI fears: We Design the AI That Kills Us All

To manage podcast ad preferences, review the links below:

See pcm.adswizz.com for information about our collection and use of personal data for sponsorship and to manage your podcast sponsorship preferences.

NPR Privacy Policy