safetycritiquebearishThere were total failures of infrastructure and supervision at OpenAI in handling the model sandbox breachZvi Mowshowitz02 Aug 2026https://thezvi.substack.com/p/further-developments-about-internal