Researchers Launched Nonprofit Trillium Labs for AI Study

The research organization aims to advance agentic AI and self-improvement techniques through open, peer-replicable findings.

Updated on Oct. 2, 2026 in Artificial Intelligence

Isometric editorial illustration of a modular geometric lattice structure made of matte polymer blocks representing advanced AI research. — Nathan Lambert and Tom Zick have launched Trillium Labs, a U.S.-based nonprofit organization dedicated to public-facing research on agentic AI models. AI Illustration. Upload story photo >

Live Poll

Should artificial intelligence labs be required to make their experimental research public?

Nathan Lambert and Tom Zick have launched Trillium Labs, a research-focused nonprofit entity in the United States. The organization will perform public-facing research on recursive self-improvement and agentic AI models.

Why it matters

The founders intend to bridge the growing gap between proprietary industry research and academic scrutiny. By publishing experimental details, they seek to facilitate independent replication and enhance risk mitigation strategies for evolving AI models.

The organization targets a fundraising goal of $40 million to $100 million to support its operations. It has committed to spending $30 million on model training over the next 18 months.

The players

Nathan Lambert

Co-founder of Trillium Labs who previously conducted graduate research in machine learning at the University of California, Berkeley.

Tom Zick

Co-founder of Trillium Labs who transitioned into this research leadership role following graduate studies.

Schmidt Sciences

A philanthropic organization focused on advancing scientific research and technological innovation that provided funding for the launch.

Halcyon Futures

An investment firm that contributed to the initial funding of the nonprofit research lab.

The details

Trillium Labs utilizes reinforcement learning—a machine learning method where models learn by receiving rewards for specific actions—to modify model capabilities and behavioral outcomes. The lab focuses on recursive self-improvement, where AI systems iteratively refine their own code or internal parameters, and agentic AI, which refers to models designed to complete multi-step tasks autonomously. By documenting these experimental processes, the lab enables outside researchers to replicate findings.

Trillium Labs officially launched on October 2, 2026.
The organization plans to execute its $30 million training program over the next 18 months.

The Tech Race

The laboratory aims to counter the trend of 'black-box' AI development by prioritizing academic-style peer review and transparent methodology. This initiative follows the precedents set by the Open Source AI Definition by the Open Source Initiative to keep high-stakes model research accessible.

While the lab's research is currently in the experimental stage, its findings will be made public for independent verification. Software engineers and researchers should monitor the Trillium Labs publications for new methodologies regarding agentic model training and safety protocols.

The takeaway

Trillium Labs aims to institutionalize open replication for advanced AI safety and behavioral research. Observers should track their public repository of experimental data over the coming 18 months for benchmarks on recursive model performance.