Human Review Halved in AI Repos: Researchers Propose "Comprehension Audits" for RSI
ronbodkin · x · 2026-10-09
A new paper by @bahradx, @ronbodkin, and Hadfield proposes "comprehension audits" as a gate on recursive self-improvement: independent auditors pick key AI R&D contributions, and the humans responsible must explain them to other humans in a thesis-defense-style process. If nobody actually understands what was built, development pauses.
Motivation: Anthropic, OpenAI, and Google each report AI writes 75%+ of new code, while human review is thinning — human review comments per line of code across open-source AI repos roughly halved between early 2024 and mid-2026.
More from coding & agent
- Devin's Security Swarm sends parallel agents to find, prove and patch exploitable vulnerabilities — DevinAI · 2026-10-09
- LangChain Podcast Digs Into Decision Models After OpenAI and Databricks Launches — LangChain · 2026-10-09
- ClickUp builds Brain² agent execution layer on E2B sandbox microVMs — badphilosopher · 2026-10-09
- Exa launches ATLAS benchmark: even priciest search agents miss ~1/3 of results — yoimnotkesku · 2026-10-09
- rea hits 21k GitHub stars: agents that reverse engineer apps down to native binaries — MatthewBerman · 2026-10-09
- PPTBench: a new benchmark testing if coding agents can rebuild visuals into editable slides — jiqizhixin · 2026-10-09