The AI Scientist-v2

A dated reading of what is claimed, reported, and independently verified in the current evidence.

Current reading

The readiness gap, in one scan

AI-assisted assembly · derived results

Evidence strength Strong

Decision

Human editorial judgment · 2026-06-28

Adopt with guardrails

Why: It is open-source, cheap per run, and has a real (if withdrawn) peer-review milestone, making it the most adoptable of the three for internal research drafting. But the makers' own quality regression, hallucinated-citation risk, and code-execution danger mean it is an assistant under supervision, not an autonomous publisher.
Next: Pilot AI-Scientist-v2 agentic tree search for narrow, computational ML write-ups inside a Docker sandbox at the documented $15-25/run budget; never auto-submit output and keep the mandatory AI-disclosure clause; route any external-facing artifact through a human gate.

Constraints

No named blocker is present in the current public projection.

Evidence summary

AI-assisted assembly

Counts and dates only. Raw signals, private excerpts, trust records, and internal corpus material are not published here.

Publication record

Initial public reading

This is the initial public reading. No earlier readiness change is recorded.