Frequently asked about medical superintelligence and Superposition Labs.
By Pablo Díaz and Víctor Perl
Co-founder, Superposition Labs, Inc.
Published
What does Superposition Labs do?
We build the deployment layer for autonomous medical AI - what we call the harness. Frontier labs build the intelligence; we build the infrastructure that lets it reach patients safely: clinical integrations, liability architecture, regulatory scaffolding, and the trust layer hospitals need before they let autonomous AI run.
What is "medical superintelligence"?
AI that matches and eventually exceeds human doctors in diagnosis, treatment planning, and clinical decision-making. Not a chatbot that summarizes notes - a system that can practice medicine. The benchmarks already show it arriving: Med-Gemini at 91.1% on MedQA, AMIE at 81.7% diagnostic accuracy against 53.3% for primary care physicians. We wrote about this in The Mountain Top.
What is "the harness"?
The infrastructure layer between a frontier medical AI model and a patient. Clinical integrations, liability frameworks, regulatory compliance, data standards, trust mechanisms - everything the AI needs around it to safely operate in a hospital or clinic. Frontier labs will not build this; it is a different business entirely. Read the full thesis in The Harness.
Is Superposition building AI models?
No. Google, OpenAI, Anthropic, and other frontier labs are building the medical AI. We build the deployment infrastructure that lets those models reach patients - the clinical integrations, the liability architecture, the regulatory scaffolding. Different problem, different company, different business model.
Is Superposition a radiology company?
No. Our first product, SignatureAPI, is document infrastructure for healthcare. Radiology is a thesis-fit future direction - imaging is where autonomous AI will likely gain clinical autonomy first - but our current work is building the deployment layer starting with the document stack.
What is SignatureAPI?
Document infrastructure for healthcare. Our first base camp: a product that solves real problems for its own users (healthcare document workflows) while quietly laying pieces of the harness in the ground. Each base camp earns its own revenue and teaches us the landscape no one has mapped yet.
When will autonomous AI be allowed to practice medicine in the US?
Utah moved first in January 2026 with HB249. We expect a patchwork of state-level frameworks through 2027, with a coherent federal framework likely between 2028 and 2030. The regulatory timeline depends on liability architecture more than on technical capability - the AI is already good enough. See our regulatory analysis.
What did Utah approve in January 2026?
HB249 authorized AI copilots that can prescribe roughly 190 chronic medications under physician supervision. The physician remains the licensee and liability holder, but the AI handles the clinical decision-making for a defined scope of conditions. It is the first statutory framework for autonomous clinical AI in the United States.
Who is liable when autonomous AI misdiagnoses?
Currently, liability defaults to the supervising clinician under respondeat superior. There is no statutory framework for AI-as-practitioner liability in any US jurisdiction. This is one of the hardest unsolved problems in medical AI deployment and a core piece of the harness we are building toward. We wrote about this in Liability Architecture.
How is clinical AI currently deployed at scale anywhere?
China. Over 260 hospitals across 93.5% of provinces run autonomous clinical AI systems backed by MIIT standards and state infrastructure. The liability model is fundamentally different - state-backed, with institutional rather than individual accountability. It is the only country with population-scale autonomous clinical AI deployment today.
Why won't frontier labs build the harness themselves?
Wrong business model. Frontier labs sell models and API access - they are not going to build hospital-by-hospital clinical integrations, negotiate liability frameworks with state medical boards, or stand up regulatory compliance infrastructure. It is the same reason AWS does not build the apps that run on it. Full argument in The Harness.
Where is Superposition based?
Between Chile and the United States. We are Chilean founders building for the US healthcare system first, with a long-term view toward global deployment. Chile gives us perspective - bimodal healthcare, first-world care for a few, developing-world access for the rest - and the US gives us the regulatory frontier. More on our About page.
Who are the founders?
Pablo Díaz and Víctor Perl. Both from Chile, both from families of physicians - surgeons, doctors, mental health professionals. Medicine was always in the room. The deployment gap between frontier medical AI and the bedside is the problem they were always going to come back to. Read more on our About page.
Is Superposition hiring?
Not broadly. We are looking for clinicians, regulators, and engineers who have been thinking about what medicine looks like after superintelligence. If that describes you, write to contact@superposition.company. We do not have open reqs on a job board - we are finding people through the work itself.
How can hospitals, clinicians, or researchers work with you?
Write to contact@superposition.company. We are particularly interested in conversations with hospital operators evaluating autonomous AI deployment, clinicians navigating the liability landscape, and researchers working on clinical integration frameworks. We respond to every message.