OpenAI chief Sam Altman says safe alignment of advanced AI remains unsolved. In the full interview Fortune published on September 15, he said OpenAI had not solved alignment and expressed his belief that other labs had not either.
Alignment here means keeping a system directed toward human intentions. Altman warns against treating cooperative behavior in today's models as assurance that the next capability level will be safe.
Before such a training run, he calls for a credible case that the models involved can be controlled and monitored. A company should pause if it cannot provide that case. His statement sets a condition for future development.
Fortune revisited the conversation on October 5. The assessment dates to September; ThePrint had already reported it on September 13. Altman declined to endorse a precise percentage for catastrophic AI risk in the interview.