Proposal: frontier AI labs should disclose and audit compute spent on recursive self-improvement

S_OhEigeartaigh · x · 2026-11-18

Nathan Calvin outlines a concrete first step for AI companies serious about pacing the frontier and preventing runaway recursive self-improvement: regularly disclose how much compute goes to AI R&D-style acceleration work, verified by a third-party auditor. Governments could mandate it, but labs need not wait — the argument is they should start voluntarily and immediately. Safety researcher SOhEigeartaigh publicly agreed.

Original post →

More from Safety

Safety channel →