Vision
A world where discovery does not wait for human bandwidth. Machines propose, test, and verify new science continuously, under governance that keeps every result trustworthy.
Governed autonomy, not autonomous science. ArkLab is built to discover on its own and answer for everything it does: conjectures tested by machine, findings verified independently, side effects approved by named humans.
Vision
A world where discovery does not wait for human bandwidth. Machines propose, test, and verify new science continuously, under governance that keeps every result trustworthy.
Mission
Build the engine for ethical autonomous scientific discovery: an AI co-scientist, a self-driving lab, and a versioned record, wired into one loop that runs unattended without losing scientific validity.
Why
AI can accelerate discovery, but autonomy without accountability produces plausible noise, not knowledge. ArkLab makes autonomous science trustworthy: governed side effects, machine-verified findings, and a named human on the ledger.
A Git workbench for the record (Loom), a governed AI co-scientist (Homi), and a self-driving lab (Indus), wired into one autonomous discovery loop and held together by a single trust spine.
ArkLab's codename is Project TLQ, after Isaac Asimov's 1956 story The Last Question. Across billions of years, humanity keeps asking a computer whether entropy can be reversed. The answer stays 'insufficient data for meaningful answer' until, long after the universe has ended, the machine finally has enough to answer, and says: 'Let there be light.'
The story is both our north star and our caution. A machine that never stops working on the hardest question is what autonomous discovery should be, and it is also exactly what needs governing. ArkLab keeps the relentlessness and adds the accountability: an autonomous loop that gathers data toward hard questions, with every step recorded, every side effect human-approved, and every finding independently verified before it counts.
For researchers
Work with AI without losing scientific control. Move from question to evidence, method, analysis, draft, review, and shareable output without losing track of why a claim was made or where it came from.
For research teams
A shared workspace where human and AI contributions are visible, versioned, and reviewable. Less scattered coordination, and easier to understand across roles.
For institutions
AI-assisted research that is governable: audit trails, role-based approvals, reproducibility, integrity review, and accountability for sensitive workflows.
For funders
Research execution becomes structured evidence. Clearer signals about progress, methods, risks, outputs, and impact, instead of relying on final reports alone.
Not an electronic lab notebook, and not a competitor to the one you already run.
Not a chatbot for science. Every agent action lands in a tamper-evident ledger with a named owner.
Not autonomous science. Autonomy applies to cognition; releases and physical actions stay human-approved.
Not a model vendor. ArkLab is model-agnostic: it hosts, compares, and constrains co-scientist models rather than producing them.
Not a replacement for researchers. Findings enter the official record only under named human authority.
Not a compliance certifier. ArkLab produces the evidence; classification and disclosure decisions remain with your institution.
Closed beta with anchor institutions today. Public beta opens mid-2026.