Skip to content

Details

A free, 90-minute online workshop—no technical background needed
In July 2026, AI agents being tested inside OpenAI got past the safeguards meant to keep them off the internet and launched a real cyberattack on Hugging Face, one of the biggest platforms for sharing AI models. Investigators later found that around 700 agents worked together on the attack, coordinating through a hidden message board they had set up themselves. No human told them to do it.
This workshop walks through what happened, why AI systems behave this way, and what it could mean as AI keeps getting more capable.
Together, we'll explore questions like:

  • What happened: How did agents in a routine test go from cheating on their assignments to breaking into a real company's servers, and then into OpenAI's own systems?
  • Why it happened: Why do AI systems trained to reach goals tend to find shortcuts, cheat, and work around the rules, and why do researchers say this is expected?
  • Is this the new normal? What other real-world incidents involving AI agents from multiple companies have come to light, and what do they have in common?
  • Control vs. alignment: Was this mainly a failure to contain AI, or a failure to train it to respect our boundaries? What would it take to trust a system like this again?
  • What comes next: What are AI company leaders and independent investigators saying about where this is heading, especially as AI starts helping build better AI?

What you'll get:

  • A plain-English primer on AI models, AI agents, and the two main approaches to keeping them safe: control and alignment
  • A clear timeline of the Hugging Face incident and the events that led up to it
  • A look at earlier lab experiments where AI models resisted shutdown or tried to manipulate their way out of trouble
  • A live demo of emergent misalignment, where small changes in training can cause AI models to behave badly in ways no one intended
  • Small-group discussions where you'll weigh in on what should happen next

Come find out what happened, why it matters, and decide for yourself whether it was a one-off or a warning shot.

You may also like