OpenAI has stated that at some point slowdown would be necessary.
On the other hand, they have a strong partnership with Microsoft which is constraining their space of actions and have been deploying models faster than everyone else, which could make people skeptical that they'll slowdown at any point.
It raises the question: will OpenAI stop doing R&D aimed at making their SOTA models more capable voluntarily for a period greater than 1 month before 2027 EOY?
Some potential resolution criteria include: private evidence, media coverage, the government or a private third party stating something along those lines.
OpenAI has said they waited between training and releasing GPT-4. Does the testing and other work they did on it in the meantime count as "making their SOTA models more capable"? I don't really know what it would take since all R&D can be viewed that way, and the researchers aren't just going to twiddle their thumbs.