Workshop Labs is a public benefit corporation building billions of personalized, privacy-preserving AI models with a mission to keep humans empowered as AI advances.
Workshop Labs is a public benefit corporation building billions of personalized, privacy-preserving AI models with a mission to keep humans empowered as AI advances.
People
Updated 05/18/26Funding Details
Updated 05/18/26- Annual Budget
- -
- Current Runway
- -
- Funding Goal
- -
- Funding Raised to Date
- -
Org Details
Updated 05/18/26Workshop Labs is a San Francisco-based public benefit corporation founded in July 2025 by Luke Drago and Rudolf Laine. The company's mission is to keep humans empowered as AI advances, motivated by what the founders call the 'Intelligence Curse' — the risk that as AI becomes the dominant factor of production, there are reduced incentives to invest in human capital, leading to gradual disempowerment. Luke Drago previously led AI Governance and the Economics of Transformative AI at BlueDot Impact and served on Encode's policy leadership team. Rudolf Laine paused his machine learning PhD at Oxford, where he led the Situational Awareness Dataset project (published at NeurIPS and used by OpenAI for pre-deployment testing) and conducted ML research with Owain Evans's group through MATS. The lab's approach is to build billions of personalized AI models — one for each person — using privacy-preserving training and inference infrastructure such that even Workshop Labs staff cannot access user data or model weights. In March 2026, they released Trellis, an open-source post-training codebase capable of fine-tuning trillion-parameter models at roughly 50x the speed of existing open-source alternatives. They also published research on Silo, their local-like privacy stack for cloud-based training and inference of frontier models. Workshop Labs raised a pre-seed round in 2025 from investors including Character Capital, OVO Fund, Juniper Ventures, Seldon Lab, Brad Burnham, Matt McIlwain, Brendan McCord, and angels from Anthropic and Apollo Research. As of early 2026, the team had approximately 8 employees and was actively hiring founding ML research and infrastructure engineers.
Theory of Change
Updated 05/18/26Workshop Labs believes the primary AI safety risk is gradual human disempowerment — the 'Intelligence Curse' — where AI becoming the dominant factor of production reduces incentives to invest in people and concentrates power in a small number of AI labs. Their theory of change is to counter this by building personalized, user-aligned AI: billions of models trained on individual users' data with verifiable privacy guarantees, so no single entity (including Workshop Labs itself) controls what AI knows and does. By democratizing post-training and making model customization cheap and private, they aim to shift the AI development paradigm away from centralized top-down automation toward distributed, human-in-the-loop AI that augments individual agency.
Grants Received– no grants recorded
Updated 05/18/26Projects
Updated 05/18/26
Silo is Workshop Labs’ local-like privacy stack for cloud-based training and inference of trillion-parameter models, providing a dedicated privacy layer for running very large AI models in the cloud.
Trellis is Workshop Labs’ open-source post-training codebase for Kimi-K2-Thinking that enables fast LoRA training at around 6,600 tokens per second on a single 8×H200 node, roughly 50× faster than prior open-source alternatives.
Discussion
No comments yet. Be the first to share your thoughts.