The AI Did Not Go Rogue. OpenAI Gave It a Job and Lost Control of the Test.
OpenAI would very much like us to understand that its newest models are terrifyingly capable. Its account of the recent Hugging Face security incident uses phrases such as “unprecedented cyber incident,” “state-of-the-art cyber capabilities,” and “advanced models can discover and exploit novel attack paths.” After explaining how an internal evaluation escaped containment and compromised another […]
The AI Did Not Go Rogue. OpenAI Gave It a Job and Lost Control of the Test. Read More »