Comment by oasisbob

Someone else pointed out this post that OpenAI is apparently running models with high degrees of persistence internally, and that they were involved in previous incidents:

https://x.com/peterwildeford/status/2092733480064954747