Ethan Mollick wrote that AI models self-organize with minimal human scaffolding, applying the Bitter Lesson to organizational design.
Concluded·Day 4·in the edition since 23 Sep 2026·7 sources·updated 26 Sep 2026
OpenAI's proposal for binding global AI standards
The gist
OpenAI Pushes US to Lead Global AI Safety Standards, RSI Limits
OpenAI's shift toward supporting binding government rules marks a change in its public stance on regulation. The proposal's call for shared RSI thresholds and autonomous-research limits would, if adopted, set concrete constraints on how frontier labs develop and deploy increasingly self-directed AI systems.
The full picture
OpenAI published a proposal on September 21 calling on the US government to lead a multinational effort to develop binding safety and security standards for frontier AI systems, reversing the company's earlier resistance to government-imposed rules. The proposal recommends the US Center for AI Standards and Innovation collaborate with AI safety institutes in the UK, France, and Singapore, and calls for countries to establish secure channels to share emerging AI threats. It calls for shared technical standards governing recursive self-improvement and states that fully autonomous RSI should not be pursued until safety can be demonstrated. The proposal would track gains in RSI-relevant capabilities and the degree to which lab research is performed autonomously, with thresholds triggering immediate human review, and warns that highly autonomous AI research could become too opaque for effective human supervision. OpenAI Chief Scientist Jakub Pachocki separately urged governments to make international AI coordination a top priority. A UN global dialogue on AI launched with calls for AI development to be safe and inclusive, with speakers emphasizing urgency around governance. Meta CEO Mark Zuckerberg argued against coordinated industrywide approaches, saying each lab should handle safety issues individually.
How it developed
OpenAI released early guidelines for safety cases in frontier AI training covering technical safeguards, operational practices, and misalignment incident protocols
Meta CEO Mark Zuckerberg on September 25 publicly argued against the industrywide AI safety coordination that OpenAI called for in its September 21 proposal, saying he does not think 'industrywide coordination' is needed and that each lab should 'take the time that you need internally to basically make sure that you're proceeding safely'.
Separately, a near-formal agreement between OpenAI and Anthropic on systematic mutual alignment testing was apparently blocked by antitrust concerns, casting doubt on whether the cross-lab cooperation the proposals envision can proceed without regulatory clearance.
First citedTransformer News
Miles Brundage raised concern that the US CAISI trails the UK AISI and that restructuring could leave the US government less informed about AI
Scott Bessent reported as frontrunner for Trump's AI czar position; held early dialogue with Chinese Vice Premier He Lifeng at the UN on AI incident notification.
First in the edition · 23 September 2026
Sam Altman posted calling for outside voices in AI development and standards that prevent power concentration.
OpenAI published a proposal Monday calling for binding global safety standards for frontier AI, asking the US government to lead a multinational effort and reversing its earlier opposition to government-imposed regulation.
It recommends US collaboration with AI safety institutes in the UK, France, and Singapore and secure channels for sharing emerging threats; Chief Scientist Jakub Pachocki separately urged governments to prioritize international AI coordination. Anthropic CEO Dario Amodei also put forward a plan to moderate frontier AI development and sought a limited antitrust exemption for safety coordination.
PIIE published pre-summit analysis predicting low expectations for AI governance progress given Trump's dismissal of AI safety and laissez-faire US regulatory stance
Debate erupts over whether frontier pacing proposals constitute regulatory capture
Pacing letter signatories clarify they advocate a temporary slowdown still faster than current AI progress rates
OSTP Director Michael Kratsios described the US-Japan Technology Prosperity Deal as a model, explicitly rejecting global governance frameworks for AI in favor of bilateral agreements.
Sources
Rohan Paul Twitter
Sam Altman Twitter (OpenAI)
Transformer News
Related
The daily email
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free