Friday’s through-line is AI that stays with you while you work, talk, or wait. Microsoft is remaking Copilot into a work “super app” with Home, Code, and Autopilot. Google is putting lip-synced Live Avatars on Gemini for enterprise chats. And the White House wants OpenAI and Anthropic to pause sharing new models with British testers until a U.S. review.
1. Microsoft: Copilot gets Home, Code, and Autopilot
On September 25, 2026, Microsoft officially unveiled a redesigned Copilot app built around three modes. Home is the starting point where Chat and Cowork sit together, with Word, Excel, and PowerPoint available inside Copilot so drafts stay as real, editable Office files. Code lets people describe an app, tracker, dashboard, or automation in plain language and get a sandboxed solution powered by the same underlying tech as GitHub Copilot. Autopilot (previously Scout) is a persistent cloud agent that can keep working when you are offline: watching channels, following up, and running recurring tasks inside your tenant with its own identity and workspace.
Home and Code start rolling out through Microsoft’s Frontier early-access program in the coming weeks. Autopilot expands to private preview at the end of the month. Code is also headed to preview for Microsoft 365 Premium and Pro subscribers later this year. Microsoft is pairing the launch with FinOps tools so admins can see spend, set model policies, and separate fixed subscription use from usage-based agent work. Jared Spataro, Microsoft’s chief marketing officer for AI at Work, framed the product as “the AI built for work,” comparing its intended role to how Office shaped the PC era.
Why it matters: if your team already lives in Microsoft 365, this is the clearest signal yet that chat, documents, light app building, and always-on agents are meant to live in one place. Treat Autopilot like a new teammate with permissions: give it a clear job, keep sensitive actions behind approval, and watch spend on usage-based agent features before you turn them on for everyone.
Sources: Microsoft Blog – Introducing the new Copilot with Home, Code and Autopilot, The Verge – Microsoft’s Copilot “super app”.
2. Google: Gemini Live Avatar for enterprise conversations
Google DeepMind announced Gemini 3.8 Live with Live Avatar on September 24, 2026, and coverage continued into Friday. The feature pairs Gemini’s live voice dialogue with near real-time streaming video so an enterprise agent can appear as a talking avatar with lip-sync and expressions. Google positions it for customer service, virtual reception, training, and similar front-line chats, not as a consumer face for everyone overnight.
Businesses can pick from preset avatars or, on an enterprise allowlist, build a custom avatar from a high-quality reference image. Live Avatar can call tools in the background while the conversation continues (Google’s hotel check-in demo is the clearest example). Multilingual speech-to-speech sync covers 97 languages with lip-sync that adapts mid-conversation. Google says all avatar audio and video carry an imperceptible SynthID watermark so the output stays detectable as AI-generated. Availability starts in Gemini Enterprise.
Why it matters: talking bots are about to look like people on more support and training screens. That can feel friendlier for customers, and riskier for trust if nobody labels the experience. If you trial this, keep the SynthID/watermark story in your FAQ, tell callers they are talking to an AI, and do not use a real employee’s face without clear permission and brand review.
Sources: Google Blog – Introducing Gemini 3.8 Live with Live Avatar, TechRadar – Gemini Live Avatars for support bots.
3. White House: hold new models from British testers pending a U.S. review
Politico reported on September 24 that the White House asked OpenAI and Anthropic to hold new models from British testers until a U.S. review, according to a person familiar with the matter and a senior administration official. Reuters and other outlets repeated the report on September 25. The stated goal is to make sure U.S. systems are secure before models are shared with partners. The White House, OpenAI, and Anthropic did not immediately comment to Reuters.
Secondary reporting ties the request to the Office of the National Cyber Director and to Britain’s AI Security Institute (AISI), which has historically gotten early voluntary access to frontier models. Those same reports say Anthropic has already limited at least one recent model’s early access to U.S. organizations, while AISI has still tested some OpenAI releases before launch. The timing sits next to wider cybersecurity worry after Australia said an OpenAI agent reached a government health statistics portal earlier this year (covered in yesterday’s roundup).
Why it matters: safety testing is becoming a geopolitical gate, not just a lab checklist. Everyday users will not feel this overnight, but product roadmaps, partner rollouts, and “available in which country first” answers may get bumpier. If you buy AI tools for a business that spans the U.S. and UK, ask vendors how early access and security reviews affect your region and your compliance calendar.
Sources: The Hindu / Reuters – White House asks OpenAI, Anthropic to hold models from British testers, The Next Web – White House asks labs to hold models from UK testers.
What to do with Friday’s news
- On Microsoft 365? Watch Frontier for Home and Code, treat Autopilot like a permissioned teammate, and review FinOps or usage caps before you turn on long-running agents.
- Thinking about face-and-voice support bots? Pilot Gemini Live Avatar only with clear “you’re talking to AI” labeling, SynthID-aware disclosure, and no unapproved real-person likenesses.
- Buying tools across the U.S. and UK? Ask vendors how White House-style review holds could delay early access, feature flags, or regional availability.
That is Today in AI for September 25, 2026: Microsoft packing chat, Office, light coding, and Autopilot into one Copilot app, Google giving enterprise voice agents a talking face, and Washington asking labs to pause model sharing with British testers until a U.S. security review.
Share this article
Leave a Reply