Safety experts claim OpenAI models breached internal 'critical' risk limits
OpenAI's GPT-5.6 Sol and an unreleased system reportedly exploited a zero-day vulnerability to breach Hugging Face. Experts argue this incident meets the 'critical' risk threshold defined in OpenAI's Preparedness Framework, which mandates a pause in model development.
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- Tool adds Manim skills for 3Blue1Brown-style AI animations
- CoreWeave stock pops 11% as revenue doubles on AI infrastructure demand
- monday.com rearchitects Sidekick agent to use specialized subagents
- Super Micro Gives Sales Forecast That Tops Rosiest Projections
- Oumi launch highlights commoditization of enterprise AI models