This Week in AI: October 12–16, 2026
Here are the biggest artificial intelligence stories from the week of October 12–16, 2026: new models, legal battles, government action and the real-world risks of AI.
Top stories
- Meta releases Muse Spark Open. Meta’s new open-weight model has 816 billion parameters, with 42 billion active per request. It has a one-million-token context window and handles text, images, video and audio. Meta calls it the most capable US open-weight model to date.
- Eight US states sue OpenAI. Arkansas, Arizona, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia allege that OpenAI misled consumers about ChatGPT’s risks and failed to protect minors. OpenAI says it has age-specific safeguards.
- The FTC investigates OpenAI and Anthropic. The US Federal Trade Commission sent both companies demands for records on security incidents, safeguards, marketing and privacy.
- Bank of America hires about 1,000 contractors to fix software flaws found by advanced AI models. They join roughly 3,000 staff already working on the fixes.
- The White House Super Intelligence task force starts work. It has 120 days to report on cyber and infrastructure risks, biosecurity and model accountability. It also plans breach-reporting rules for AI labs.
New models and products
- Google Gemini 4 Argon became broadly available in the Gemini API and the Gemini app for paid subscribers.
- Microsoft MAI-Base-2: a model with 2.8 trillion parameters, of which 80 billion are active. Microsoft says it trained far more efficiently than rival open models. Microsoft also announced its Maia 300 AI chip, with production starting in 2027.
- NVIDIA Nemotron 3.5 Super VL: an open multimodal model built for AI agents.
- OpenAI GPT-Live-1.1: better voice conversations in ChatGPT, with new voices and improved accents.
- Anthropic Claude Opus 5.5 fast mode: up to 2.5x faster responses, now open to all developers.
- Meta Muse Wonder: new models aimed at agentic work and coding.
- Google Universal Cart: a shopping cart in Search and Gemini that combines products from many stores, launched in the US.
- Mistral Timestral: small time-series forecasting models that run on ordinary CPUs.
Safety, security and society
- About 420 researchers signed an open letter asking labs to keep models with advanced hacking abilities offline during early testing.
- Anthropic’s research team warned that small AI models can already carry out many steps of copying themselves onto cheap hardware.
- An Italian bank lost €95 million to fraudsters who used an AI clone of a CEO’s voice. About €53 million has been recovered.
- The US Federal Reserve’s Beige Book found that firms in 5 of 12 districts were hiring less or cutting staff after investing in AI.
- Anthropic, OpenAI and the Simons Foundation committed $15 million to help mathematicians adapt to AI.