Morning digest: Claude Code, Cursor apps, Anthropic IPO

Morning digest for Friday, Nov 6. Claude Code was putting some users' account email addresses into the User-Agent headers of web requests it made, and Anthropic says a fix is shipping. Cursor can now build and publish live web apps from a cloud agent. Investors cited by the Wall Street Journal say Anthropic's planned IPO may need a lower valuation after this week's AI stock selloff. Also today: Meta's Muse gets real-time voice and expressive avatars in the U.S., Perplexity launches always-on Teammates for Max subscribers, Google flips K–12 AI tools off by default, and Anthropic's biology lab used Claude to map the first vault particle found in a prokaryote.

1. Claude Code was putting some users' account email addresses into the User-Agent headers of outbound web requests, and Anthropic says it will remove the email from the model's context next week. A User-Agent header is the short label a browser or script sends to identify itself to a website. In GitHub issue 107757, filed Wednesday, a macOS user on Claude Code 2.1.318 said the tool "divulged my email to arbitrary 3rd parties in the User-Agent request header while conducting web searches," and that they were getting "dozens of scam emails per day." Anthropic engineer Lydia Hallie wrote on Hacker News that the company put the account email into the model's context so Claude would stop inventing names for things like @author tags, but "when Claude wrote a command itself, like a curl call, it sometimes put your email address in the User-Agent as a contact address." "I'm sorry we let this happen," she wrote. "We're removing the account email address from the context Claude Code gives the model" in a release "early next week." Until then, versions "2.1.317 and later have a built-in mod (on by default)" called cc-plugin-privacy; "The stable channel is still on 2.1.311, so update to 2.1.317 or later." The mod's own README says, "This is a courtesy, not a boundary," and that it only checks certain shell calls that pair the address with a User-Agent flag—not WebFetch, WebSearch, MCP tools, scripts, or values built at runtime. Some users on 2.1.318 say the mod is missing or still lets matching commands through. The reporter's scam flood is their own account; Anthropic has confirmed the email-in-User-Agent behavior, not every spam claim. What this means for you: If you use Claude Code, update past 2.1.317, open /plugins and leave cc-plugin-privacy on, and treat anything still in the agent's context (git config, secrets, personal details) as able to leave the machine on a tool call. This follows yesterday's digest note on cheaper Claude Code subagent status updates.

2. Cursor can now build a live web app from a prompt, set up a database and sign-in, run security checks, and give you a shareable link. In a Nov. 5 changelog, Cursor says, "you can build live apps directly in Cursor and share them with a link," and that a cloud agent will "build the app, set up a database, configure auth, and deploy it." You start a cloud agent, say you want to build an app, pick a domain when you publish, and "Nothing goes live until the checks pass and you explicitly publish the app." Team admins can see every app and secret, add their own review rules, and unpublish. On individual plans you can share apps publicly; team apps sit behind Cursor sign-in. "This capability is now available on all paid plans with cloud agents enabled." Try it: Update Cursor, start a cloud agent, and ask it to build a small internal tool (a signup form, desk map, or simple dashboard). Review the security findings before you hit publish.

3. Investors planning to buy into Anthropic's IPO say the company may need a lower valuation than the roughly $1.5 trillion figure that had been expected, the Wall Street Journal reports. According to Investing.com's write-up of the Journal and Gulf News, Anthropic has been aiming to list the week before Thanksgiving (Gulf News says the week of Nov. 16) and to raise up to $100 billion. Three major investors who plan to take part told the Journal the valuation "will likely need to come down," with one saying it could be "closer to $1 trillion," after this week's AI-related stock selloff. Investing.com, citing the Journal, says Anthropic had about $146.6 billion in cash as of Sept. 30, and that executives and advisers plan to meet in the coming days; delay or a smaller raise are also options under discussion. Anthropic did not immediately comment in the Gulf News report. These are secondhand accounts of a gated Journal story, not an Anthropic filing. What to watch: Whether Anthropic prices near $1.5 trillion, cuts the target, shrinks the raise, or slips the date—and how that landing affects the rest of the AI funding trade. Related reading from yesterday: METR's report on AI accelerating Anthropic's own research.

4. Meta is rolling out real-time voice chat and expressive avatars for Muse in the United States. In a Nov. 5 post, Meta says voice and expressive avatars are "rolling out now in the Muse app in the US to start," on mobile, web, and Mac—tap the phone icon in the top-right corner to start. The avatar "reacts as it listens and speaks, with expressions and gestures that match the conversation," and you can choose voice-only, avatar beside the chat, or full screen. Meta says there are "hundreds of voices to choose from," and "You can even interrupt if you need to." Voice on AI glasses and Muse Charm is "coming soon." Try it: If you're in the U.S. with the Muse app, tap the phone icon and try a short brainstorm or trip-planning talk-through; skip it if you don't want another always-listening assistant on your phone.

5. Perplexity launched Teammates, always-on work agents for Max subscribers. In its blog post, Perplexity says Teammates "build on the power of Computer" and can "plan the work, check results along the way, and bring completed work back for review," remember project goals across runs, monitor ongoing work, pass tasks to each other, and join Slack. "Teammates are initially launching on the web and soon on iOS for Perplexity Max subscribers." The Verge reports Max is "$200 / month." Free and Pro users can upgrade under Settings → Account → Subscription. What to weigh: At Max pricing this is a power-user / team tool, not a casual upgrade. If you already live in Perplexity Computer, Teammates are the always-on layer; if you don't, the $200/month seat is the gate.

6. Google is turning Gemini, Gemini Notebook, Gemini in Classroom, and Google Search off by default for K–12 student accounts until school administrators turn each one on. In a Google Keyword post, education VP Chris Phillips says access to those products "will be entirely off for K-12 student accounts until school administrators explicitly turn them on," with more detail for schools "in the coming months ahead of the changes rolling out." Google is also adding Guided Learning (step-by-step coaching instead of straight answers) in Classroom globally now, and piloting it in Search at select U.S. schools later this month. Ars Technica reports Google clarified that schools can leave Gemini apps off while enabling Search—but Search still includes AI Overviews / AI Mode, so there is no Google option for student search with AI fully stripped out. What this means for you: If you work with schools or kids' Google accounts, watch for admin notices before the change lands; districts that want search without AI will need another search provider, not a Google toggle.

7. Anthropic's biology lab used Claude to help find and map the first 3D structure of a vault particle from a prokaryote—a simple cell such as a bacterium—or, in this case, a bacteriophage that builds the vault inside one. Vaults are large, hollow protein shells whose job in cells is still unknown. In a research post and bioRxiv preprint, Anthropic says the phage protein L05 forms a vault from 96 copies of itself (vs. 78 in mammals) and "encloses about 25% more volume" despite being shorter. With Claude's help, a purification that "conventionally takes three days" took "about 90 minutes," and "a first full model was ready in 34 hours." Anthropic is clear that Claude also made mistakes—including a bad growth medium that set the project back two weeks—and that the team has not yet shown these vaults form in the wild hosts, or what they do. What to watch: Whether outside labs can reproduce the L05 structure and whether anyone finds a clear function—the post is a research demo, not a product launch.