Latest / AI World Today: Daily News / Claude Duet Works While You're Away | AI World Today: Daily News #4
Transcript
- Hi, and welcome. This is AI World Today: Daily News for Friday, October twenty-third, twenty twenty-six. Today, Claude Duet wants to keep your work moving after you close your laptop. Google is bringing voice-powered document creation to more workplaces. ElevenLabs is taking live translation into phone calls. And an update on an attack involving a cloud backup provider shows why connected services need careful oversight. Let's start with Claude Duet. Anthropic introduced Duet this week as a more proactive assistant. It has its own computer and browser in the cloud, so tasks you delegate can keep running when your phone or laptop is off. The more interesting part is how it finds work to bring back to you. With the right accounts connected, Duet can check for updates every few hours, prepare a morning briefing, and assemble drafts or research for your review. Imagine opening your day with the background for an upcoming meeting already gathered, and a draft response to an email that needs your attention. That's the kind of workflow Anthropic is aiming for. The appeal is easy to understand. Much of the work around a task happens before the task itself: finding the right thread, remembering what was agreed, and getting the relevant information into one place. A useful assistant could take some of that preparation off your plate. Anthropic says Duet is designed to ask for approval before sending messages, making payments, signing documents, or deleting anything. You should still read what it prepares. A draft based on a misunderstood email can look perfectly polished. There are also some practical limits. Duet is an invitation-based research preview for Pro and Max subscribers in the United States. Joining the waitlist does not guarantee immediate access, and Team and Enterprise accounts are not included in this preview. During the preview, there is no additional subscription charge. But background work counts toward the same usage limits as the rest of Claude. An assistant that keeps working while you're away can also use up more of your allowance. And those checks happen periodically. An urgent message could arrive between them. For me, the interesting test is whether Duet consistently returns with something useful enough to save time. A concise briefing and a good draft would be a strong start. A pile of material that needs extensive correction would be another job to manage. Next, Google wants you to talk your next document into shape. The feature is called Docs Live, and this week's update brings it to eligible business editions of Google Workspace on mobile. Google first introduced the experience for its paid consumer AI plans in September. The news now is its wider availability at work. The workflow starts with a conversation. You explain the document you need, and Gemini proposes an outline. You can adjust that plan before it turns into a draft, then use your voice to request changes. For example, imagine finishing a project meeting and telling your phone: put together an update for the team, separate the decisions from the open questions, and finish with the next steps. That example describes the intended workflow. The quality of the result will depend on the information you provide and the material the system can access. Docs Live can use context from sources such as Drive, Gmail and Chat, as well as the web. Source settings let you control which material it uses. That makes the outline an important checkpoint. You can see whether the document is heading in the right direction before asking for the full version. The current voice experience is English only, and access depends on having an eligible paid plan. It is available through the mobile Docs app on Android phones and iPhones. For anyone who finds it easier to explain an idea aloud than to start with an empty page, there is a clear attraction here. Before sharing the result, check the names, decisions and deadlines against the original material. Staying with voice, ElevenLabs has announced a service for translating conversations as they happen. The idea is that each participant speaks their own language and hears the other person's words translated. Think of a Spanish-speaking customer calling an English-speaking support team about a delayed delivery. In a service that has integrated the technology, translation could sit between the two sides of that conversation. ElevenLabs says its Realtime Translation product supports spoken output in more than ninety languages. It can also stream translated text and transcripts, which could be used for captions. The initial release is an alpha for enterprise customers using the company's developer platform. Phone networks, contact centers and other providers need to build it into their services. Your ordinary phone calls do not automatically gain translation because the product has been announced. Deutsche Telekom is already working with ElevenLabs on calling features for its Magenta AI Call Assistant. There is a useful detail about how this works. Providers can tune how soon the translation starts. Starting earlier reduces the wait, while allowing more of a sentence to arrive can improve the context available for translation. That is a real design choice in a live conversation, where people pause, interrupt each other and change direction. The translated speech also uses a selected output voice. The system does not automatically reproduce the original speaker's exact voice in another language. The potential is considerable for everyday support calls. The next thing to watch is how well these integrations handle accents, interruptions and important details in normal conversations. Finally, an update on AI-assisted cyberattacks. Microsoft reported this week that a group it tracks as Storm zero nine four zero has been using an AI framework called Artex to automate parts of its operations. Microsoft assesses the group as aligned with the Chinese state. The report links the group to the compromise of application credentials associated with ConnectWise Cloud Backup, previously known as SkyKick. Those credentials mattered because customers had already granted the backup application access to their Microsoft three sixty-five environments. A trusted service connection became a route attackers could attempt to use. Microsoft says attempted sign-ins reached more than two thousand two hundred and fifty customer environments. Most of that targeting was unsuccessful. An attempted login is not evidence that an organization lost data. Researchers at Wiz also described related activity against software and technology companies. They saw attack steps chained together, with later requests adapting to earlier responses. They said most of the attempts they examined failed. The concern is the ability to repeat and adjust familiar techniques at scale. There is a fresh operational update today. ConnectWise says restoration testing and security checks are continuing through the weekend. Subject to successful validation, it expects to begin enabling access to restore data from existing backups early next week. It is telling affected customers to keep application connections disabled until they receive the instructions for their service. The practical lesson reaches beyond this incident. When reviewing access to business systems, include the permissions held by connected applications. Those connections can remain powerful even when no employee is actively signing in. That's it for today. Thanks for listening. If you enjoyed the episode, follow AI World Today: Daily News to keep up with what's happening in AI. You'll find links to the stories and sources in the description, and more news and analysis at AI World Today dot com. This episode is narrated by an AI voice. See you tomorrow.