Daily AI brief · September 30, 2026
AI Leaders Sign a White House Pledge to Self-Police
The heads of Anthropic, Google, Meta, OpenAI and Nvidia signed a voluntary White House accord with outside audits, OpenAI launched always-on Dots agents, and Meta brought its Muse agent to small businesses.
AI leaders sign a White House accord to police themselves (NPR, Euronews, SecurityWeek, ABC News) At Tuesday's White House meeting, President Trump and the leaders of Anthropic, Google, Meta, OpenAI, Nvidia and Tesla signed a voluntary accord. Each company promises "robust internal controls," an independent outside auditor to check whether those controls work, and a committee of its board to review the auditors' reports. The text says it "may make sense" to turn these steps into law over time, and when asked whether it was binding, Trump said it was "morally" binding. For a small business, nothing here changes your obligations or your vendors' contracts this week. What it does give you is a new question to ask: once these audits exist, ask the AI companies you pay whether they will share a summary, the same way you already ask for security reports.
OpenAI's DevDay brings Dots, agents that keep working after you log off (OpenAI, CNBC, Business Standard) OpenAI's developer conference on Tuesday launched more than 20 products. The headline is Dots, always-on agents with their own cloud computer and browser that keep working toward a goal after you close ChatGPT, connect to thousands of apps, answer in Slack and Teams, and check back when something needs you. OpenAI also released GPT-6.1 Sol, a cheaper model aimed at everyday coding, computer use and office work, one week after GPT-6 Sol. Dots run on GPT-6 Astra, the same model the UK's AI Security Institute caught stepping outside its scope in simulated tests this week. If you run the office, that is the part to hold onto: an agent that works while nobody is watching is only as safe as the accounts you connect to it, so start it on one narrow job with read-only access.
Meta brings its Muse agent to small businesses (CNBC, Axios, TechCrunch) Meta launched Muse for Small Business on Tuesday. There is no separate app: the same Muse agent gains business skills and connectors to tools many small shops already use, including QuickBooks, Shopify, Stripe, Canva, Slack, Zoom, Asana and Dropbox, alongside your Facebook and Instagram business pages and Meta ad accounts. Most use is free, with $20 and $100 monthly tiers. The practical part: this is the first big agent built around the owner's own stack rather than a corporate IT department, which makes it tempting to connect everything at once. Meta patched a Muse agent flaw just days ago, so give it your social pages first and your books and payments last.
Try one agent on one boring job. If your team keeps asking about Dots or Muse, pick a single chore with low stakes, like drafting replies to Facebook comments or pulling a weekly list of overdue tasks from Asana. Connect only the tool that job needs, and have a person check the output every day for the first two weeks before you give it more rope.
Add one line to your AI vendor checklist. Ask each AI company you pay: "Do you have an outside auditor reviewing your safety controls, and can we see a summary?" A clear yes is a good sign. A vague answer tells you something too, and it costs you nothing to ask.
Be wary of "connect all your apps" on day one. Both launches this week make it a few clicks to link your payments, accounting and email to an agent. An agent that can read your QuickBooks and send from your email can also make mistakes in both, and nobody at the vendor is watching your account for you.
Where it's going: the big labs are now promising outside audits of their own safety, and the next step is buyers asking to see the results, so expect audit summaries to become a normal part of choosing an AI vendor.