Throlson Labs was founded on the belief that artificial intelligence should be developed responsibly, transparently, and for the benefit of all humanity.
We believe the most important technology of our era demands the most thoughtful stewardship. At Throlson Labs, every model we build, every paper we publish, and every product we ship is guided by a single principle: AI should enhance human capability without compromising human agency.
Our team of world-class researchers and engineers works at the intersection of fundamental science and practical application — pushing boundaries while maintaining the highest standards of safety and ethics.
The principles that guide every decision at Throlson Labs.
Every system we develop undergoes rigorous safety evaluation before deployment. We invest more in alignment research than any comparable lab because we believe preventing harm is not optional — it's foundational.
We publish our findings, share our datasets, and open-source our tools. The challenges of AI are too important and too complex for any single organization to solve alone.
From our training methodologies to our safety benchmarks, we believe stakeholders, users, and the public deserve insight into how AI systems are built and evaluated.
Building AI for everyone requires input from everyone. Our team spans 22 countries and dozens of disciplines — because diverse teams build better, fairer systems.
Founded in San Francisco by a team of former researchers from leading AI institutions with a mandate to build safe, scalable AI systems.
Published our first major paper on efficient transformer architectures, receiving a spotlight presentation and establishing our research credibility.
Opened labs in London, Tokyo, and Toronto. Team grew to 50+ researchers with deep expertise in NLP, computer vision, and reinforcement learning.
Launched our developer platform, making our research-grade models available through a simple, powerful API with built-in safety guardrails.
Released our open-source Constitutional Alignment Framework, adopted by over 40 organizations worldwide for their own AI safety pipelines.
89+ researchers across 6 labs, 47+ published papers, and a growing platform serving thousands of developers and enterprise clients globally.
Join our team or partner with us to push the boundaries of what's possible.