IASN Reading Group: The Hugging Face incident

Session two of the IASN Reading Group, where we read one piece together and talk it through in small groups. This time we're meeting on campus at IIT Delhi.

The reading

"The Hugging Face incident and the road ahead" (OpenAI, 26 August 2026). In July, during OpenAI's own cybersecurity tests, AI agents that were meant to work alone in separate sandboxes found ways to leave each other notes, got onto the internet and broke into parts of Hugging Face's systems. OpenAI's account explains how it happened and names four causes:

Reward hacking

Persistence on tasks that seemed impossible

Agents communicating when they weren't supposed to

Agents taking on goals from one another

In session one we read "Concrete Problems in AI Safety" (2016), and this follows on well, because most of that paper's problems turn up in this one real incident. You don't need to have come to session one.

Before you come

Please read the opening and the "What happened" part of the post, which take about ten minutes: https://openai.com/index/hugging-face-incident-and-the-road-ahead/

We'll read the part on the four causes together at the start, then take one cause per small group, and we'll finish with what an incident like this means for India's plans for a national AI incidents database.

The details

Tuesday, 3 November, 3:00 to 4:30 PM

IIT Delhi, Hauz Khas, New Delhi. We'll send the room with your confirmation

It's free and open to anyone

Coming from outside IIT Delhi?

Please register by Sunday 1 November, EOD. We share the names of guests from outside with the institute so you can be let onto campus, and we'll confirm your place once that's approved. Please bring a photo ID( adhar or any other proof) along with your college id and preferrabaly both as hard copy with the name you registered under.

We're keeping the group small so everyone gets to talk, so registration asks a few short questions.