OpenAI asks US to lead global standards for self‑improving AI
The company said fully autonomous recursive self-improvement is not happening today and should not be pursued unless it can be done safely.
OpenAI called on Monday for the United States to lead an effort to write international technical standards for AI systems that help build the systems that follow them. The company set the proposal out in a post on its own site.
Reuters, Politico, the Financial Times and the Wall Street Journal each reported the call, which OpenAI made in a post titled "Building standards for the next phase of AI". Techmeme, citing Axios, said the post came ahead of an address by Sam Altman at the United Nations.
The position is OpenAI's own and concerns a practice OpenAI expects to use. Techmeme quoted the post as saying that automated research could improve alignment, but that "fully autonomous RSI is not happening today" and should not be pursued unless it can be done safely.
Two accounts that read the post described what OpenAI wants built. AndrewCurran_ said OpenAI proposed linking national AI safety institutes into one network and tracking how much research inside a lab is being done by AI. The account choblin29 said OpenAI wants the same measurement of RSI-relevant progress.
The same two accounts said OpenAI drew a limit around its proposal. choblin29 said the post argues the standards should not become model licences, mandatory pre-release reviews or government approval requirements. Neither the wire reports nor the accounts recorded a response from any US agency.