Machines
Alignment
Closing the gap between what capable models are asked to do and what they actually do, and keeping it correctable.
Substrate is an independent research lab
studying the structure of intelligence.
Machines, molecules, and the systems that shape our world.
Explore the research →
Closing the gap between what capable models are asked to do and what they actually do, and keeping it correctable.
Reverse-engineering the features and circuits behind model behavior, so internals can be checked, not assumed.
Probabilistic methods for molecular systems, from nucleation pathways to the behavior of biological compounds.
How deployment and governance choices shape the real-world impact of capable AI.
Our researchers have worked across:
We work with the bodies writing the rules for AI and security, at the state, federal, and international level. Our aim is narrow and practical: turn vague legal standards into tests that can actually be run.
Pre-rulemaking comment on the Automated Decision-Making Technology Act (SB 26-189) and the Chatbot Safety Act (HB 26-1263), proposing measurable standards for the statutes' vague operative terms.
Comment on the proposed Policy Statement on the Suppression of Accuracy in AI Systems (Docket FTC-2026-0859), arguing that calibration and honest abstention are themselves forms of accuracy, and that the deception test should rest on measurement.
A Statement on AI's Transformation of the Economy, convened by the Stanford Digital Economy Lab.
An open letter supporting mandatory screening and recordkeeping for synthetic nucleic acids, to keep pace with AI-enabled biological risk.
Contributors work alongside existing commitments. Credentials give context; we care most about reasoning on open problems.
Independent and volunteer-run. One address for everything, press, partnerships, positions, and support (compute, grants, infrastructure, advisory time).