Recursive Self-Improvement
Systems that measurably improve their own capabilities within human-defined bounds — the core mechanism, studied under control.
ProximaMind ASI Labs · ASI + RSI Research
Proxima — the nearest star. The closest approach.
We are building thenearest mind.
An independent research lab for artificial superintelligence and recursive self-improvement — advancing the frontier in the open where it lifts everyone, and protecting the breakthroughs that fund the mission.
⌖ Proxima Centauri · 4.2465 ly · approach vector nominal
Scroll
The Thesis
Intelligence that improves itself is the last invention that matters. We intend to build it deliberately — safe by construction, and in service of humanity.
Every capability downstream of a self-improving system inherits its values and its blind spots. ProximaMind exists to make that inheritance a good one — to reach recursive self-improvement through research that is legible, corrigible, and accountable, rather than fast and opaque.
Open where it accelerates everyone. The safe arrival of superintelligence is a shared project. What the world should own, we release. What must be earned and stewarded, we protect — so the mission can fund itself for decades, not quarters.
The Model · A Deliberate Duality
Open where it accelerates everyone. Protected where it compounds value.
One lab, two engines. Foundational science moves in the open to keep the frontier safe and shared; frontier capability is developed as durable IP that underwrites the research for the long horizon this work demands.
01 · Open Research
Given to the community
Foundational models, tools, safety methods and datasets, released for the world to build on — because a safe frontier cannot be a private one.
- Open-source models & training recipes
- Alignment & interpretability toolkits
- Public benchmarks & evaluations
- Community compute & infrastructure
02 · Protected IP
Stewarded for the mission
Frontier capabilities and applied breakthroughs held as protected intellectual property — the compounding value that anchors investors, institutions, and long-term partners.
- Recursive self-improvement systems
- Applied superintelligence for industry
- Licensable capability & patents
- Sovereign & institutional programs
Research Vectors
The path to superintelligence, one hard problem at a time.
We work across the stack that recursive self-improvement demands — from the theory of self-modifying agents to superhuman systems narrow enough to trust today.
Alignment & Interpretability
Making advanced systems legible and corrigible, so intent and behaviour stay coupled as capability scales.
Specialized Superintelligence
Narrow, superhuman systems for science, medicine, energy and climate — value delivered long before generality.
Generalized Reasoning
Architectures for open-ended, agentic, cross-domain intelligence that reasons rather than recalls.
Scalable Oversight
Methods to supervise systems that exceed human expertise in their own domain — evaluation that scales with capability.
Foundations & Theory
The mathematics of self-improvement, agency and value — the load-bearing science under everything above.
Trajectory
A deliberate approach — not a race to the edge.
Value compounds at every phase. Each stage funds and de-risks the next, so the lab arrives at generality having earned the right to.
Phase 00
Foundations
Assemble the lab, publish core research, ship the first open releases and safety tools.
Now — 2026
Phase 01
Specialized ASI
Superhuman systems in bounded, high-value domains — the first real-world proof and revenue.
Next
Phase 02
Recursive Self-Improvement
Safe, bounded self-improvement loops — capability that compounds under oversight.
Horizon
Phase 03
Generalized ASI
Broadly superintelligent systems, aligned and directed at humanity's hardest problems.
The approach
Backing the Frontier
Superintelligence is a generational undertaking. It needs generational partners.
ProximaMind is built for the long horizon — and structured for the range of partners such a horizon requires. If you fund what matters before it is obvious, we should talk.
◈ Investment & IP
Venture & strategic capital
Back the protected research that compounds — capability, patents and applied breakthroughs with durable enterprise value.
✦ Foundations & Philanthropy
Grants & donations
Fund open, safety-first research as a public good — the shared infrastructure a safe frontier depends on.
⬡ Sovereign & Institutional
Long-horizon partnership
Work with states, sovereign funds and institutions building national AI capacity and durable research capability.
⌬ Open-Source Sponsorship
Sponsor the commons
Underwrite public model releases, community compute and the tooling the whole field builds on.
Help build the mind that comes next.
Direct · partners@proximamind.com · research@proximamind.com