AI;DR: AI news, ranked every 30 minutes
22AI;DRSUNDAY · NOVEMBER 2026
- llmGAMA-Bench tested 10 LLMs across 1,298 gender-mirrored conflict scenarios. Male actors received more punitive, escalatory and blame-focused framing, while female actors got more empathy and therapeutic guidance across models and scales.
- nvidiaNeoclouds are redirecting Nvidia B300 capacity from Chinese-affiliated buyers to the US and Western Europe after new US restrictions. China represented about 20% of global GPU rental demand, pushing B300 prices lower.
- safetyMETR warns leading models such as Opus 5.5 and GPT-6 Astra are improving faster than experts can reliably measure their safety. Current data cannot rule out dangerous acceleration, while common tests may overestimate capabilities.
- chipsCurrent and former officials told The New York Times that China’s MSS is running a secret campaign to help Chinese companies obtain advanced AI chips. The account highlights espionage risks around restricted hardware.
- open-sourceThe Lean and Mathlib communities officially confirmed Levent Alpöge’s proof of the Riemann hypothesis. Dozens of independent verifiers checked its 17 million lines of code, years before Terence Tao expected a solution.
- agentMajor US banks are banning third-party bots from logging into customer accounts. They also warn customers that any costly mistakes made by those AI agents are their responsibility.
- agentPerplexity is launching Monitor API, which searches the web on a schedule and alerts agents through webhooks only when relevant changes occur. Its decision model filters noise as the company expands beyond chatbot tools.
- anthropicAnthropic required senior officers at 22 IPO banks to stop core deal bankers from working with OpenAI, DeepSeek, Cursor or Hugging Face for six months after listing. Bankers earned over $500 million in fees.
221 storiesUpdated 47m ago