About
Readback Labs is a nonprofit research lab that measures how people oversee AI systems. The question of whether humans still understand what AI does for them is mostly answered by assertion. We think it deserves evidence.
noun · radio procedure
Repeating an instruction back to the sender so both sides know it was understood.
Pilots read a clearance back so the controller can hear that it was understood. The habit exists because a confident "roger" is not the same as understanding, and the gap between the two costs lives.
AI systems now hand work to people all day, in the same confident tone. Readback Labs measures whether the people on the receiving end are still reading it back: checking, questioning and correcting what they are given, or waving it through.
Origins
Readback Labs, Inc. was incorporated in 2026 as a California nonprofit public benefit corporation. It measures how people oversee AI and how fast AI is changing the work it is used for, in collaboration with partners, publishes its approach, and shares the results to inform decision-makers.
The lab is small by design: a founder and executive director, and a team of contributors and part-time volunteers who do the engineering, research and outreach.
Agenda
Measurements of AI capability, control and human oversight that are validated against people and published with their sources, so that claims in either direction can be checked.
Practices that ask the people responsible for AI systems to show they understand what those systems did before the systems are given more autonomy, the way a readback confirms a clearance.
A world where AI developers' safety claims are checked by qualified outside organizations: building the demand for that work and the competence to do it well.
Putting what we learn in front of civil society, standards bodies, agencies and companies as capabilities accelerate, early enough for it to matter.
People
Founder and executive director
Founder and executive director of Readback Labs and lead author of Comprehension Audits to Mitigate Risks from Automated AI Research. He has founded three companies, led AI engineering at the Vector Institute, served in the CTO office at Google Cloud, and was lead author of Canada's national AI compute strategy.
Engineering, analysis and outreach
Builds the lab's data pipelines and tooling, works on the analysis, and handles outreach. Previously a data analyst at Tandem PV. Northwestern University, 2025.
Outreach · part-time volunteer
An experienced executive, Sachin leads outreach to fintech pilot partners. [Details to come]
Student contributor · Purdue University
Contributes to the engineering and research of the lab's projects as a part-time volunteer.
Student contributor · Princeton University
Contributes to the engineering and research of the lab's projects as a part-time volunteer.
Contact
Questions about the research, contributing, the survey, or what the evidence does and does not show. We read everything, and we answer the serious ones.