Ex-DeepMind researchers launch nonprofit to keep humans in AI safety loop

A group of former Google Alphabet Inc. ↗ 0.84% DeepMind researchers has launched a nonprofit called Sampura Research, dedicated to ensuring that humans remain central to the oversight of increasingly powerful AI systems. The London-based organization, formally announced on August 24, aims to build what it calls stronger "judges" — hybrid human-AI systems capable of evaluating whether models behave safely and as intended.bloomberg+2

A Counterpoint to AI-Supervising-AI

Sampura Research was co-founded by Rishub Jain, Joshua Jacob, and Alex Adams, all of whom previously worked at Google DeepMind. Jain spent seven years at the lab, including two years focused on scalable oversight research and contributions to projects such as AlphaFold. Jacob served as a co-lead for human data engineering at DeepMind and previously worked at Waymo.EdTech Innovation Hub+4

The nonprofit positions itself as a counterpoint to the prevailing industry approach of using AI to supervise AI — a method that offers speed but, as Jain argues, still leaves gaps that human judgment can help fill. According to Bloomberg, Jain believes people retain an essential role in verifying model behavior, even as AI systems become more capable.Bloomberg Law+1

Building the Infrastructure for Oversight

Sampura's research agenda centers on developing datasets that measure how effectively humans and AI systems can verify model outputs, then using those benchmarks to improve oversight methods. The organization defines a "judge" as "a human, AI, or hybrid system that assesses correctness and alignment of an AI's behavior in a conversation or agent trajectory".sampura+1

Planned workstreams include forward-looking harm datasets, methods for allocating subtasks between people and AI, and an open-source human rating platform. The team is recruiting at least six experienced AI researchers in London, with salaries ranging from £100,000 to £290,000.EdTech Innovation Hub+1

Amid a Broader Safety Reckoning

The launch comes during a period of heightened concern about AI oversight capacity. A Reuters report noted that AI firms "can't yet contain what they've built", while the nonprofit METR — which independently evaluates frontier models for OpenAI, Anthropic, Google, and Meta ↘ 1.64% — has struggled to hire enough staff despite offering salaries above $500,000. In late July, more than 1,100 AI practitioners signed an open letter urging the U.S. government to support an international mechanism for slowing frontier development when necessary.news.aibase.com+2

Jain's departure was part of a broader talent exodus from DeepMind in June 2026, during which the lab lost five core researchers in six days.kucoin.com