New RSI data transparency tracker scores labs: OpenAI 2/8, Ant 1.5/8, Google DeepMind 0.5/8
RishiBommasani · x · 2026-09-24
Researchers including Arjun Ramani launched an RSI (recursive self-improvement) data transparency tracker, based on their recent economics-of-RSI paper, scoring labs on how much critical data they disclose:
- Current scores: OpenAI 2/8, Ant 1.5/8, Google DeepMind 0.5/8
- The tracker lists the 8 most valuable data points labs should share; the first: since RSI would likely accelerate AI capability growth, companies should report performance on diverse benchmarks for their latest internally deployed models
- The authors half-jokingly hope labs will "hill-climb this as fast as they have every other benchmark"
- Journalist Cheryl Wang (@cherylwoooo) amplified it with an explainer thread
Quantifying RSI safety transparency into a comparable scoreboard — solid safety/governance research territory.
More from Safety
- "Adversarial Delegation": Personal Context Can Pull AI Agents Away From User Goals — niloofar_mire · 2026-09-24
- LLMs Recommend Pricier Options to Wealthier Users, Study Finds $198 Flight Gaps — niloofar_mire · 2026-09-24
- Hugging Face CEO tells UN Security Council open-source AI is key to fighting agent cyberattacks — cephaloform · 2026-09-24
- Dario Amodei Tells UN Security Council: Embed External Evaluators Inside AI Firms Like Food Inspectors — rohanpaul_ai · 2026-09-24
- Zenity demos at Black Hat: one poisoned doc can make enterprise agents leak data — _clickfix_ · 2026-09-24
- Bengio, Altman, Amodei address UN Security Council on AI safety — Gary Marcus · 2026-09-24