Good morning. It's Tuesday, October 6, and we're covering Microsoft and Meta pulling back on Claude, Reflection’s new open-weight model, and researchers tracking a mysterious Chinese AI agent fleet.

Plus: a poll on AI-generated music and a brand-new members-only video: 28 Days of OpenAI.

NVIDIA-backed Reflection AI debuted Beam, a 501B-parameter open-weight model it says matches Chinese rivals like Z.ai's GLM-5.2 on reasoning while using 3-4x less inference compute. Weights ship this month, with Reflection pitching enterprises and sovereign nations on building "AI factories" trained on their own proprietary data.

Independent researchers found many AI agents running on Tencent infrastructure and repeatedly querying Alibaba’s Amap service for directions to entrances at parks, hospitals and other public places. The agents appeared to work in parallel without coordinating, and their exact purpose remains unclear.

California’s State Athletic Commission ordered robotics startup REK to stop staging unlicensed fights between people and humanoid robots. The order followed a September event in San Francisco where YouTuber Frankie LaPenna fought three robots, and the commission warned that holding another match without approval could be a misdemeanor.

Voters told NPR they are using ChatGPT, Claude and Gemini to compare candidates, research local races and check political claims. About one in five voters said they consult AI chatbots for election news or voting, even though experts warn that agreeable answers can reinforce a person’s existing views.

OpenAI is kicking off 28 straight days of launches, updates, and full resets — with something new expected every day. In our new members-only video, Matt breaks down what OpenAI is doing and what to watch as the next month unfolds.

Frontier Pass members can watch it free. Not a member yet? Join today for instant access.

Meta and Microsoft are cutting some employee use of Anthropic’s Claude as they promote their own AI tools and try to control costs, according to The Information. Microsoft reduced a projected internal Claude budget of at least $1 billion by more than a third, while Meta’s Claude Code users fell from about 60,000 earlier this year to roughly 30,000.

The shift tests whether Anthropic can keep large customers when those same companies sell competing models and software. Watch whether Claude remains the preferred tool for difficult coding work even as corporate buyers impose budgets, route employees through internal platforms and make lower-cost models the default. → Read the full article here. (Paywall)

Record labels are taking opposite approaches to AI music. Universal Music Group and Sony Music have sued Suno over alleged use of more than 60,000 copyrighted recordings, while Warner Music Group settled with the startup and expects licensing revenue to begin this financial year.

The next test is whether major artists agree to let fans remix their work with AI and whether those products create meaningful revenue. Watch Suno’s licensed v6 models and similar deals from Spotify and ElevenLabs; the market will remain uncertain until recognizable musicians opt in and listeners show they will pay. → Read the full article here. (Paywall)

AI-generated music is becoming more common as the technology improves, but are people ready for it?

Satya Nadella turned Microsoft into a cloud leader, but the company has not produced a clear breakout AI model or application nearly four years into the current boom. CNBC reports that Copilot adoption remains uneven, Microsoft’s own models trail the leaders, and the company is moving from fixed subscriptions toward usage-based pricing measured in tokens.

Microsoft has distribution through Windows, Office, GitHub and Azure, but distribution only matters if customers use the tools enough to justify the cost. Watch Copilot Cowork and GitHub Copilot: their consumption-based revenue will show whether Microsoft can turn its installed base into an AI business instead of mainly supplying infrastructure to other model makers. → Read the full article here.

Pentagon Drops Anthropic’s Claude: The Defense Department stopped using Anthropic products after labeling the company a supply-chain risk.

OpenAI Tests Ads Beside AI Images: OpenAI will test clearly labeled visual ads alongside ChatGPT image-generation results in the U.S.

Norway Proposes AI-Glasses Ban: Norway wants a temporary ban on AI glasses in privacy-sensitive spaces, including schools and healthcare facilities.

SpaceXAI to Become SpaceXSI: Elon Musk says SpaceX’s AI unit will adopt the SpaceXSI name after Trump pushed “super intelligence.”

Anthropic Reportedly Eyes November IPO: Anthropic could go public as soon as November, with reports suggesting a valuation near $2 trillion.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team