Let’s talk

ANIFOLD · BUILDING SAFE RSI SYSTEMS

Let AI evolveSafely

Data × Infra × Safety for Autonomous AI Improvement

Learn from real tasks. Turn each discovery into a verified next step

Scroll into the next iteration
Scroll into the next iteration
0108

01 / WHAT WE BUILD

ANIFOLD

Data×Infra×Safety

Data: Find what to learn next

Infra: Keep learning and updates continuous

Safety: Verify new capabilities before the next iteration

02 / THE ANIFOLD LOOPDiscover · Learn · Verify · Return

Discover, learn Verify, return

  1. 01/ 06INFRA

    Meet the real world

    Agents interact with environments and carry out tasks. Their trajectories provide the evidence for the next iteration

    What moves forwardAgent trajectories

  2. 02/ 06DATA

    Find the capability gap

    Evaluate what happened, identify failures, and determine which capability needs attention next

    What moves forwardTargeted capability gaps

  3. 03/ 06DATA

    Create the data that matters

    Turn identified gaps into focused training data, grounded in what the agent actually needs to improve

    What moves forwardHigh-value training data

  4. 04/ 06INFRA

    Bring learning back to the model

    Connect new data to training and model updates through continuous learning infrastructure

    What moves forwardAn updated model

  5. 05/ 06SAFETY

    Verify before the next step

    Evaluate capabilities and risks against deployment constraints. An update that fails verification returns for improvement

    Verified → ReturnNot yet → Improve
  6. 06/ 06INFRA

    Return. Then improve again

    A verified update re-enters the environment. New experience becomes the starting point for the next improvement

    What moves forwardThe next iteration

Scroll to explore · Swipe to changeEvery return passes through safety verification

+03 — THE PEOPLE

Built by people who know the frontier

Continuous learning infrastructure and AI safety, bringing systems engineering and frontier research together for safe AI improvement

01

Xingjun Ma

Chief Scientist · AI Safety

Making stronger AI stay safe

Associate Professor and PhD Supervisor at Fudan University. National-level Young Talent, Shanghai Leading Talent, and AI100 Young Pioneer. Author of 100+ papers with 18,000+ Google Scholar citations, specializing in LLM and agent safety, with research cited by the International AI Safety Report and European national security agencies

Fudan University

02

Mark Jiang

Co-founder, Infra

Keeping AI improvement running

Former OpenAI infrastructure expert and Google algorithm engineer, with experience building infrastructure for continuous frontier-model iteration and large-scale AI systems. PhD from the University of Melbourne, with academic experience at UCLA and Stanford

OpenAI · Google · University of Melbourne

Meet the core team

BUILDING SAFE RSI SYSTEMS

Build the
next iteration

Data × Infra × Safety

[email protected]