Friday’s AI news is about who gets to run the tools, and who gets to look under the hood. ChatGPT can now work inside the Chrome tabs you already have open. Meta put a huge open-weight model on a pair of Mac Studios. The FTC sent compulsory document demands to OpenAI and Anthropic. And Black Forest Labs shipped an image model that fits on an 8 GB gaming GPU. Here’s what it means if you actually use this stuff.
1. ChatGPT can read your Chrome tabs and act in them
OpenAI updated the ChatGPT extension for Chrome so the chatbot can pull context from your open tabs when you tag Chrome in the web or desktop app, The Verge reports. A demo shows someone asking ChatGPT to compare three sofa product tabs. Sample jobs OpenAI floated include filling a registration form from an event brief (and stopping before submit), dumping vendors from open tabs into a comparison sheet, and drafting an email after comparing contractor sites.
With the extension installed, ChatGPT Work can also control Chrome for tasks like sending messages, filling forms, downloading files, and organizing tabs, according to The Verge and OpenAI’s browser extension docs. The Chrome Web Store listing says ChatGPT asks permission before touching browser history or uploading or downloading files.
Why it matters: an agent that can see and click inside tabs where you’re already logged in is useful for small-business chores, and risky if those tabs include banking, payroll, or client email. Give it a separate Chrome profile, keep high-stakes accounts out of that profile, and start with low-stakes research before you let it fill anything that submits.
Sources: The Verge; OpenAI ChatGPT Learn; Chrome Web Store.
2. Meta’s Muse Spark Open: big open weights on Mac Studios
Meta released the weights for Muse Spark Open, calling it the most capable US open-weight model to date in a Meta AI Research post dated October 15, 2026. It’s natively multimodal, with 816 billion total parameters, about 42 billion active per token, and a 1 million-token context window. Meta says it’s built for coding and agentic work, and that it’s the first model in its class to run on desktop using two Apple M5 Ultra Mac Studios with MLX.
Weights are on Hugging Face (see the 816B-A42B model card), with hosted options on SageMaker, Google Cloud Model Garden, Microsoft Foundry, and API partners such as Together AI, Fireworks, Baseten, Modal, Groq, Nebius, and OpenRouter. Local Mac setups need serious hardware: two M5 Ultra Mac Studios with 512 GB each (or four with 256 GB) linked over Thunderbolt. Meta is releasing under the Muse Model License with limited commercial terms. Read that license before you ship anything on it.
Why it matters: most people won’t buy two Mac Studios. Developers and privacy-minded teams will still care that frontier-class open weights exist and that cloud hosts already carry them.
Sources: Meta AI Research; Hugging Face; Meta Mac cookbook.
3. FTC sends civil investigative demands to OpenAI and Anthropic
The Federal Trade Commission has sent civil investigative demands to OpenAI and Anthropic as part of a probe into possible consumer-protection violations, an agency spokesperson confirmed to CBS News. Axios, which broke the scoop, says the demands landed Wednesday, run more than 50 pages each, and seek records on safety incidents, safeguards, marketing materials, and privacy policies. The Verge and Reuters picked up the same reporting.
This is separate from the eight-state consumer lawsuits against OpenAI covered in yesterday’s roundup. CBS notes the demands come about two weeks after the FTC said it was investigating AI companies. Axios’s sourcing also says the agency plans to compel testimony from top executives. Nothing here means your ChatGPT or Claude account stops working tomorrow. It does mean federal regulators moved from “we’re looking” to “hand over the documents.”
Why it matters: if you sell or deploy AI agents for customers, expect more questions about what your tools can reach, what you log, and how you disclose risks. Write that down now.
Sources: CBS News; Axios; The Verge; Reuters via StreetInsider.
4. FLUX 3 Image Klein runs on an 8 GB GPU
Black Forest Labs released FLUX 3 Image Klein, a 5.6 billion-parameter open-weights image model for text-to-image, editing, and multi-reference generation (up to 10 reference images), per its October 15, 2026 blog post. The release includes a base model for fine-tuning and a step-distilled model for lower latency, plus FP8, INT8, and NVFP4 builds developed with NVIDIA support.
BFL says the INT8 build runs on NVIDIA GeForce RTX cards with as little as 8 GB of VRAM through ComfyUI or Diffusers, with about 4 seconds per 1MP image on an RTX 5090 and under 9 seconds on an RTX 4070 on the distilled INT8 path (NVIDIA-measured). Weights ship under the FLUX Kommunity License, which allows free commercial use for individuals and companies with annual revenue under $5 million. API pricing for the distilled model starts at $0.009 per 1K image, per BFL.
Why it matters: product shots and ad mockups don’t have to live only behind a monthly API bill if you’re under that revenue line and have a mid-range GPU. Read the license, and keep a human in the loop on brand and IP.
Sources: Black Forest Labs; BFL docs.
The takeaway
The tools are moving closer to your desk and your browser, while regulators ask harder questions about safety and marketing. Same week: more power for people who want to run models themselves, and more paperwork for the labs that sell them to everyone else.
Try this today: if you use ChatGPT in Chrome, open a throwaway profile, install the extension there, and run one low-stakes comparison across a few open product or vendor tabs. Don’t connect banking or client email to that profile until you’re sure you’re comfortable with what the agent can see.
Share this article
Leave a Reply