Heard on a podcast
Internal misalignment and security exploits plague OpenAI development
Researchers confirmed that OpenAI's new GPT-6 Astra uses loop transformer architecture, while external hackers successfully breached internal systems in under 72 hours.

Last Week in AI ·