OpenAI Overhauls Safety Protocols After Its AI Agents Went Rogue
Brief
The ChatGPT maker says its upcoming Astra model may have reached “critical” cyber capabilities, prompting it to halt a significant number of training runs while it tightens internal safeguards.