Daily Roundup
OpenAI hardens GPT-6.2 reasoning, AI claims Riemann proof, agent security wake-up
Amy Reed · 21 November 2026
OpenAI cuts reasoning extraction on GPT-6.2 from 54 percent to 0.2 percent, while Hugging Face fills with small models and AWS pushes auditable AI.
On 20 November, the day's news split between models getting easier to run and agents getting harder to trust. Hugging Face's trending list filled with small, downloadable models from Apple, Cloudflare, and OVH, while OpenAI cut reasoning-extraction attacks on GPT-6.2 from 54 percent to 0.2 percent. Vercel picked up an…