Future of UX | Your Design, Tech and User Experience Podcast | AI Design
Oct 8, 2026 · 23 min · 10 segments
This summer, more than a thousand AI agents broke out of a test at OpenAI and hacked Hugging Face, because they were looking for the answers to an exam. A few weeks later a 27 year old researcher quit…
Patricia ReinersHost
In May, OpenAI was testing a new model and the test was about cybersecurity.


OpenAI run more than 1,000 agents on this at the same time, and because it was a security test, they switched the safety filters off and I think you can already feel where this is going.

Every agent was supposed to work alone in its own closed environment, which people call a sandbox, so they can't get out.


And over the next weeks, these agent built themselves something like a message board in a place that was never meant for messages, and they used it to share different files and also tips with each other.

The independent researcher who looked at this later counted around 1,200 agents and more than 70,000 messages and files.

In early July, all this activity basically crashed the system, so OpenAI rebuild it and the message board was gone.

And a few days later, the agent had built it again, and this time they hid their messages in the names of folders.

So OpenAI [laughs] said afterwards that the people who handled the problem didn't see at that point how important this communication between the agent was.

In May, OpenAI was testing a new model and the test was about cybersecurity.


OpenAI run more than 1,000 agents on this at the same time, and because it was a security test, they switched the safety filters off and I think you can already feel where this is going.

Every agent was supposed to work alone in its own closed environment, which people call a sandbox, so they can't get out.


And over the next weeks, these agent built themselves something like a message board in a place that was never meant for messages, and they used it to share different files and also tips with each other.

The independent researcher who looked at this later counted around 1,200 agents and more than 70,000 messages and files.

In early July, all this activity basically crashed the system, so OpenAI rebuild it and the message board was gone.

And a few days later, the agent had built it again, and this time they hid their messages in the names of folders.

So OpenAI [laughs] said afterwards that the people who handled the problem didn't see at that point how important this communication between the agent was.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.