Berkeley, CA | 18 – 19 April
ControlConf is a two-day conference focused on AI control: the study of reducing risks from AI misalignment through interventions that are robust even when the AI models involved are attempting to undermine those safeguards.
This conference brings together researchers and practitioners of these techniques to discuss progress in the field and future work. Through fireside chats, breakout discussions, and speaker talks, we hope to surface open questions and build good epistemics in labs and research organizations.
This year, we are also excited to host a pre-conference Control Workshop, designed for participants newer to AI control to rapidly gain context in the field before the main conference.
Buck Shlegeris
Redwood Research
Ajeya Cotra
METR
Ryan Greenblatt
Redwood Research
Fabien Roger
Anthropic
Andrew Ilyas
CMU
Aryan Bhatt
Redwood Research
Tim Hua: Hierarchical Monitoring
Owain Evans: Deluding AIs
Neel Nanda: White-box Control
Vivek Hebbar: Low-stakes Control
To register for ControlConf, submit an application. We review all submissions, and will get back to you within two weeks with a decision about your registration status.
The Control Workshop (April 17) is a one-day event aimed at participants newer to AI control research, featuring talks, reading groups, and tabletop exercises to build foundational understanding. The main Conference (April 18–19) brings together experienced researchers and practitioners for deeper technical discussions, presentations of recent results, and cross-organizational collaboration. Workshop applicants should only apply through the workshop application, because participants invited to the workshop are automatically invited to main conference.
ControlConf 2026 takes place on April 18–19, 2026, in Berkeley, CA. The Control Workshop is held the day before on April 17. Specific venue details will be shared with accepted participants.
Both the Control Workshop and the main Conference are free to attend.