RSI Progress

Tracking AI’s progress toward recursive self-improvement

What has been measured, what changed, and what the evidence tells us about recursive self-improvement.

Progress toward self-improving AI

Selected site evidence · Summary reviewed 27 September 2026

AI is taking on useful parts of AI development. Systems can investigate research problems, run experiments, improve training methods and optimize code. Some improvements have already been used in real model training.
Progress is uneven. The reports show gains on research tasks and a growing ability to handle longer tasks, but also experiments that stall or fail. Stronger benchmark scores do not automatically translate into faster real-world work.
People still guide the process. AI can lead parts of research, while people set goals, evaluation standards and resource limits. The amount of independent control varies across systems and research areas.
Sustained self-improvement remains unresolved. Some agents retain useful changes and reuse modified versions of themselves. The tracked evidence has not established a reliable process in which successive generations also become better at finding and producing further improvements.

Where are we? · 2026-09-10

Assessment from the paper by Duan et al. (Sep. 2026) ↗

Science

Embodied intelligence

Software engineering

Healthcare

DemonstratedEarly evidenceNot established

Levels describe different kinds of autonomy, not necessarily a required sequence.

What each level means

B0 Refine output: Changes refine an output during the current task or session. Future independent tasks inherit no accepted system update.
L1 Execute improvements: Humans prescribe the target, update procedure and acceptance criteria. AI executes the procedure and accepted changes persist into later tasks or rounds.
L2 Choose how to improve: AI diagnoses weaknesses and chooses interventions and experiments. Humans continue to set objectives, task boundaries and evaluation criteria.
L3 Choose what to learn: AI chooses or generates subsequent learning experience based on the evolving learner’s state. What it learns from changes as the learner changes.
L4 Adapt from deployment: Ongoing deployment or environmental interaction determines persistent changes in memory, skills, code, harnesses or parameters reused on later operational tasks.
L5 Improve the improvement mechanism: The procedure governing future improvement is itself revised, retained and invoked in subsequent improvement rounds. It may be an improver, search or research policy, evaluator or successor generator.

The top rung, L5: AI improves how it improves

Structural L5Bounded prototypes and emerging research or industrial systems.

Effective L5Some bounded meta-improvement and transfer; reliable multi-generation accumulation remains unresolved.

Recursive accelerationNot established by this survey’s reviewed evidence under comparable resources.

No single score
No single benchmark or percentage captures progress toward RSI.
Results keep their context
Each score belongs to one system, test version, and setup.
Capability ≠ autonomy
Doing research tasks well is not the same as controlling the improvement loop.