Artificial Intelligence · Projected milestone
Controlled recursive self-improvement
45% by 2038 · including earlier years
The year is an editorial timing estimate. The percentage applies to occurrence by the stated deadline. These estimates are not calibrated odds.
AI systems improve their own designs under verified constraints.
Explore this milestone on the interactive mapWhat the forecast means
Milestone stage: demonstration
unknown
End-to-end autonomous successor improvement within explicit operational constraints.
Completion criteria: Three successive model generations improve on pre-registered held-out tasks without humans designing the model changes; disclose compute, rejected candidates and proof assumptions. Each run is constrained by verified access/resource boundaries. This resolves bounded controlled improvement, not proof of benevolent goals.
This page describes a forecast, rather than an achieved result. Read the methodology and assumptions.
Built on
Led to
Sources & evidence
- Research acceleration: The view inside OpenAI 2026-09-06
OpenAI reports an automated research intern, increasing agent contribution and substantial human intervention on longer tasks; it targets an automated researcher in March 2028. Internal measurements are preliminary, and compute and controls still constrain work.
- Measurements for understanding the pace of AI development inside frontier labs 2026-09-17
Anthropic reports AI leading 26% of R&D in August, no fully autonomous measured R&D subset, and about 30,000 concurrent internal agents. It reports complete action-monitor coverage, which is not complete detection. The measurements use internal/model judgments.
- Demonstrating Reinforcement Learning and Run Time Assurance for Spacecraft Inspection Using UAVs 2024-05-10
The authors demonstrate reinforcement learning with runtime assurance on a physical UAV inspection experiment. Guarantees depend on a specified model and envelope; this is not a universal proof of AI alignment.
Record reviewed 2026-10-18