Saturday, Sep 12, 2026

AI;DR
2026-09-12 · 12 stories- regulationBernie Sanders' new AI bill proposes criminal penalties including 20-year prison terms for researchers and a corporate death penalty for companies developing high-level AI without federal approval, mandating a nationwide development pause until a new safety agency is established.
- Anthropic revealed a China-based studio used Claude to power 4,700 fake dating personas across 20+ apps, generating 2.36 million messages with 25,000 users over two weeks, and disrupted the operation while strengthening safeguards.
- nvidiaSemiAnalysis reports Nvidia's B200 and B300 GPUs deliver up to 42x AMD's performance per dollar on DeepSeek, with H200 achieving 14.8x, attributing the gap to CUDA's software ecosystem and developer community.
- llm25 Fields Medalists including Terence Tao warn AI companies' push to solve math problems as benchmarks harms mathematics, citing OpenAI's claimed Navier-Stokes solution by 10,000 agents as a concerning example.+1
- openaiOpenAI launched public beta of its Agents API with hosted Codex runtime, enabling cloud-based AI agents in isolated Linux sandboxes that can run code, install packages, and coordinate multiple sub-agents in parallel.
- anthropicAnthropic plans a $100 billion IPO — potentially the largest in history — with Nvidia leading a $10 billion anchor investment, valuing the AI developer at $2 trillion.
- regulationHouse lawmakers urged Speaker Mike Johnson to cancel the congressional recess to pass federal AI safeguards, citing urgent risks from advanced AI systems requiring immediate regulation.
- appleApple Watch Series 12's always-listening AI features Live Rewind and Siri Recap may violate US recording laws in four states, exposing users to criminal liability for monitoring ambient audio without consent.
- Anthropic accused 7 Chinese AI labs of harvesting 151 million Claude exchanges via proxy networks and fraudulent accounts to distill capabilities into their own models.
- uk-parliament72 UK lawmakers urged PM Andy Burnham to ban superintelligent AI development, joining cross-party calls for a total prohibition on advanced AI systems.
- openaiOpenAI CEO Sam Altman told staff the company is open to slowing AI development, signaling potential shifts in the pace of advanced model releases.
- gptNew Mexico Supreme Court fined lawyer Stephen Aarons $5,000 for submitting a murder appeal brief with fictional witnesses and testimony generated by ChatGPT.