Q&A with Aryan Bhatt (Redwood Research)

What does it actually take to keep a powerful AI system safe if it’s trying to get around you?

Join us for a live Q&A with Aryan Bhatt, Senior Member of Technical Staff at Redwood Research, who’ll be joining us virtually.

Redwood is a Berkeley-based AI safety nonprofit, best known for starting the field of AI control: building safeguards that still work even when a model is actively trying to get past them. Their work includes the Alignment Faking paper with Anthropic and some of the most-cited research on monitoring and catching misbehaving AI agents.

Aryan leads Redwood’s high-stakes control team, which studies how to catch an AI agent before it does something catastrophic. He got into the field through MATS, where he worked on LLM agent evaluations with METR. He now mentors MATS scholars himself. Before all that, he studied math and CS at Hunter College.

He’ll talk about how he got into AI safety, what research at Redwood looks like day to day, and how students can get involved, and then take your questions.

Location: Siebel 0218 / 0220 (Aryan joins by video)
Food: Free pizza and drinks
Length: About an hour. No AI background needed.