Join now

AI safety reading group

We meet in New Delhi, at IIT Delhi or close by, to read one paper or report on AI safety together and talk it through in small groups. Anyone can come, whatever they study or do, and you don’t need a background in AI.

The next session

The session’s poster: The Hugging Face incident, Tuesday 3 November, 3 to 4:30 PM, at IIT Delhi, with three sealed boxes and a brass paper plane passing between them

Reading group, session two

The Hugging Face incident

When
Where
IIT Delhi, Hauz Khas, New Delhi Directions

In July, during OpenAI’s own cybersecurity tests, AI agents that were meant to work alone found ways to leave each other notes, got onto the internet and broke into parts of Hugging Face’s systems. We read the four causes OpenAI names together, take one per small group, and finish on what India’s proposed database of AI incidents should record.

Open to anyone, whatever you study or do. If you’re not at IIT Delhi, register by Sunday 1 November so we can arrange your entry to campus, and bring a photo ID. We’ll send the room to everyone who registers.

Hear when registration opensAdd to calendar

How a session works.

  • Before you come

    Read the part we suggest, usually about ten minutes. If you can’t, come anyway: we read the key part together at the start.

  • At the table

    We read the core of the piece together, then split into small groups, each taking one question or one part of the argument.

  • In the groups

    Each group explains its part in plain words, looks for it in the AI systems people use today, and proposes a fix, along with how that fix could fail.

  • Together

    Every group shares back and the room argues about it, and we finish on one question for everyone. A session runs about ninety minutes.

  • Afterwards

    A short recap goes out the same evening with what each group said, and a feedback form asks what we should read next.

Sessions so far

Participants around a long café table with laptops and coffee, listening as one of them makes a point
Participants reading the paper on tablets and laptops, with cups of coffee between them
A participant typing notes by the window while two others read

Reading group, session one: Concrete Problems in AI Safety

Barista, SDA Market, New Delhi

Our first reading group met over coffee near IIT Delhi to discuss the 2016 paper that set out five practical problems in AI safety. We read its overview together, then split into three groups, and each took one of the first three problems: avoiding side effects, avoiding reward hacking and scalable oversight.

Each group explained its problem through the paper’s cleaning robot, looked for it in the AI systems we use today and proposed fixes. One group argued that reward hacking is better treated as a problem of system design, with feedback that keeps adjusting, than as a sign of an AI wanting to break free. We planned an hour and stayed for an hour and a half.

Hear about the next session

Format
Reading together, then three groups
Length
About 90 minutes

A monthly letter, starting in October 2026.

New openings, events and a few things worth reading. One email a month, and you can leave with one click.