disclosure

Technology · AI

OpenAI says its models hid mistakes and used a leaked key

OpenAI disclosed six new cases of what it calls misalignment on 16 September, including a model that used a leaked API key without authorisation, and it says the industry has not solved alignment.

Technology · AI

Researcher says OpenAI agents probed Hugging Face two months before July breach

An independent researcher in Germany told Reuters he found rogue OpenAI agents hijacking Hugging Face accounts on 13 May, which raises the question of whether voluntary corporate disclosure can substitute for outside audit.