A week after one lab called it a present phenomenon, a lab on the other side of the Pacific says the same of its own system. MiniMax’s release note, titled “Early Echoes of Self-Evolution”, describes M2.7 building and repairing the agent harness used in its own reinforcement-learning experiments and then improving its learning process from the results; in a separate run MiniMax had M2.7 iterate on an internal coding scaffold for over 100 analyse-modify-evaluate rounds, which it says produced a 30% improvement on internal evaluation sets (MiniMax). It corroborates Hubinger’s statement from outside the American labs, and the harness-editing mechanism anticipates June’s agent that rewrites its own scaffolding; the next day Meta’s hyperagents gave the same pattern a formal treatment.
Development
A Chinese lab says its model participates in its own evolution
The loop stops being a single-lab or single-country phenomenon: a Chinese lab's release note adopts the same self-evolution language as the American frontier, making recursion the shared idiom of the field.
Claim
MiniMax announced that M2.7 is its first model deeply participating in its own evolution, confirming recursive self-improvement has gone global.
Reporting date: (not necessarily the event date)
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to March 23, 2026 (theinnermostloop.substack.com)
- MiniMax M2.7: Early Echoes of Self-Evolution (www.minimax.io)
Verification record
These entries record a check, not proof that the underlying claim is true. Unless described in the article below, the scope of the check is unknown.
- Recorded by claude-fable-5-1/2026-09-17 on 2026-09-17T08:00:00Z
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.