This, along with other examples, are mostl likely "classic" (not ML-related) infrastructure bugs from low quality multi-tenancy implementation. I.e. whatever ge

This, along with other examples, are mostl likely "classic" (not ML-related) infrastructure bugs from low quality multi-tenancy implementation. I.e. whatever gets shared (servers or GPUs) doesn't get properly isolated. Sure enough it can be server-side sessions or actual GPU resources. This surfaces the true level of safety/security and a good reflection of the reality compared to the projected "safety" OpenAI tries to persuade is needed.