Sakana AI's "The AI Scientist"

An automated-research system built by Sakana AI that generates, runs, and writes up its own machine-learning experiments, with a reviewer-graded novelty score as its fitness function and no cross-run memory. Its single AI-authored result — a workshop paper scored 6/7/6 at the ICLR 2025 ICBINB workshop — was withdrawn before publication by the team's own prior agreement, a fact the field's coverage routinely drops in favor of a later, human-written Nature paper describing the system (myth-ai-scientist-nature-paper).

Matters to this vault as the field's reigning cautionary case for ungated autonomous science, and as the explicit foil the broader automated-science lineage's newer entrants measure themselves against: the 2026 NV-diamond multi-agent paper that prompted this hub's creation names Sakana's AI Scientist (critiqued in Beel, Kan & Baumgart 2025 for reporting performance improvements based on flawed experiments) as one of two prior systems its own self-critical design is built in reaction to.

References

• myth-ai-scientist-nature-paper
• claim-agent-memory-field-shifted-storage-to-experience-2026
• comparison-seek-among-peers-2026-07
• Capture: 10-inbox/raw/2026-11-01-outside-arxiv261040589-collaborative-ai-agents-for-autonomous-experimental.md