Re:Felony Bench
by Grombobulous on Saturday August 22, 2026 @04:42PM
(#49403697)
That’s all dependent on believing the premise of the story given by OpenAI.
Was it really a training run in a VM? Was it really only intended to touch an internal repository? How do we/HF know and verify that?
I agree with you that it’s unlikely, but it’s not out of the realm of possibility. Corporations have intentionally committed felonies many times.
As far as I recall the vendor of the repository patched the exploited vulnerability, so that checks out.
Considering that there would be no clear benefit to such criminal behavior, in my opinion "unlikely" is putting it mildly.
All so that OpenAI could do a bit of bragging and massively delay their own work and planned model rollouts?
That seems extremely unlikely.