Here is today's AI news. Story 1. Meta. Meta unveiled Muse Charm, a pocket-sized AI assistant device powered by its Muse agent, at Meta Connect on September 23.. Meta unveiled Muse Charm at Connect on Sept. 23, a pocket-sized device for using its Muse AI assistant without opening a phone app. It has a roughly 2-inch screen, fingerprint-activated voice access and built-in 5G, but can’t make calls. The device signals Meta’s push to make AI agents accessible through dedicated hardware, beyond phones and smart glasses. CEO Mark Zuckerberg said it packs Muse’s “real-time voice and avatar stack”; Meta aims to ship it by December, with no price announced. Story 2. OpenAI. OpenAI announced its always-on Dots agents, which can pursue user goals across apps, at its September 29 DevDay event.. OpenAI unveiled Dots, always-on AI agents, at its September 29 DevDay, aiming to move AI from answering prompts to pursuing ongoing work. Powered by GPT-6 Astra, each Dot runs on its own cloud computer, learns from feedback and can connect to more than 4,000 apps. Users can reach them through ChatGPT, Slack and Teams; custom rules and approval requirements govern sensitive actions. The move intensifies competition in autonomous workplace AI, including with Meta. Story 3. OpenAI. OpenAI released GPT‑6.1 Sol, a new multimodal model, on September 29.. OpenAI introduced GPT-6.1 Sol on Sept. 29, an upgrade aimed at coding agents, computer use and professional workflows. The company says it nearly matches flagship GPT-6 Astra on those tasks at one-fifth its standard API token prices: $2 per million input tokens and $10 per million output tokens. The launch intensifies competition to deliver capable AI agents at lower operating costs; Sol is available in ChatGPT Work, Codex and the API, but not regular Chat. Story 4. AI companies. OpenAI, Anthropic, Google, Meta, Nvidia and xAI executives announced a voluntary frontier-AI safety accord with the White House on September 29.. On Sept. 29, OpenAI, Anthropic, Google, Meta, Nvidia and xAI leaders signed a voluntary White House accord on frontier-AI safety. It calls for controls during training and deployment, internal monitoring teams, independent external audits and board oversight, targeting risks including cyber-, bio- and chemical threats. The pact could set shared industry practices, but it is not legally binding; the accord says measures might later be codified into law or regulation. Story 5. AI industry. A study presented at a Brookings Institution conference warned that the AI infrastructure buildout could require more than $10 trillion through 2032 and bring wider financial risks.. A Brookings conference paper by Columbia economist Stijn Van Nieuwerburgh estimates U.S. AI infrastructure investment will total $10.3 trillion from 2025 to 2032—about 3.6% of GDP annually—funding data centers, power, networks and chips. The scale exceeds past major U.S. infrastructure booms. As tech firms turn to private credit, leases and off-balance-sheet vehicles, opaque debt could spread losses if AI revenues disappoint. The paper estimates the buildout would add 183 gigawatts of data-center capacity by 2032. That's all for today's AI news briefing.