Last checked November 21, 2026 · News roundup
Today in brief: AI agents that click and type on your computer went from demo to plan feature this week. Gemini can now operate a Mac for paying US subscribers, and Anthropic opened a computer use beta for Claude Team and Enterprise workspaces. ChatGPT also removed its 10-minute dictation cap, Canva is adding a switch to turn its AI off, and an Anthropic mathematician says Claude helped him prove the Riemann hypothesis, a claim no one has confirmed yet. In business news, ANT shares slid back toward the $83 IPO price on day two, Salesforce is reportedly cutting jobs again before its December 2 earnings, and Inception will switch off its Mercury API on December 28.
The nine stories at a glance
| Story | Date | Who it affects | Cost |
|---|---|---|---|
| Gemini computer use on Mac | Nov 19 | US, English, AI Plus / Pro / Ultra | Included in those plans |
| Claude computer use (beta) | Nov 20 | Team and Enterprise workspaces | No extra fee mentioned |
| ChatGPT Live Dictation | Nov 18 | All ChatGPT plans | Free feature |
| Canva AI off switch | Rolling out | Canva Free and Pro | Free setting |
| Meta Muse data mix-ups | Nov 18–19 | Muse users, tested by staff | Muse Charm reportedly $300 |
| Riemann hypothesis claim | Nov 20 | Researchers, Claude users | Unverified |
| ANT stock day two | Nov 20 | Investors | About $86 at midday |
| Salesforce job cuts | Nov 17–20 | Staff, CRM buyers | Earnings Dec 2 |
| Inception Mercury API ends | Dec 28 | Developers on Mercury | Migration needed |
Part 1 · AI apps and models
Gemini can now use your Mac: plans, limits and what it does
The short version: Google AI Plus, Pro and Ultra subscribers in the US can let Gemini operate their Mac, in English only for now.
Google announced computer use for the Gemini desktop app on macOS on Thursday, November 19, according to TechCrunch. Once you give it an instruction, Gemini works through native Mac apps, local files and websites in the background, and you watch a live preview inside the chat while your own screen and cursor stay free. Google's examples include filling out forms, repeating the same task across a batch of files, sorting files into folders by rules you set, and building slideshows from files or online dashboards.
You can pause or stop the agent and approve its actions before they happen. Google did not say when the feature will reach other countries, languages or Windows. Apple, meanwhile, is hiring an engineering manager for an "AI agent that can use a computer," Bloomberg reported, and is eyeing a launch as soon as the end of next year. The same report says macOS 27.3 will warn users before they give third-party agents full access to their Mac.
Your move: start with a low-stakes chore such as renaming a folder of files, keep approvals on, and watch the first few runs before you hand it anything involving payments or work accounts.
Claude computer use for Team and Enterprise: how admins turn it on
The short version: Claude can now see the screen and work in desktop apps for Team and Enterprise customers, but the beta is off until an owner switches it on.
Anthropic's release notes list computer use in beta for Team and Enterprise plans from November 20, rolling out to organizations gradually. It runs in the new Claude experience in Claude Desktop on macOS 14 or later and on Windows, and people can also start tasks from Claude on the web or mobile while Claude Desktop stays open on their computer. The release note mentions no extra charge.
Only an Owner or Primary Owner can enable it: go to Organization settings, then Capabilities, check that "Chat and Cowork unified" is on, and flip the Computer use toggle under Claude Desktop. Admins can block specific apps or allow only a list, and each user still turns it on per computer and approves each app. On a Mac it needs Accessibility and Screen Recording permissions. It is not available for organizations using the HIPAA-ready setup, customer-managed encryption keys, FERPA/K-12 settings or Zero Data Retention for Claude Code. Anthropic warns there is no sandbox between Claude and your apps, and screenshots are saved with the chat. Switching it off can take about an hour to reach every desktop.
Your move: treat it like any new integration, and run it through the same review you would use to vet a Claude skill before your team installs it. Start with an "allow only these apps" list.
Desktop AI agents this week: who can use what
Gemini for Mac
Status: live, US only, English
Plans: Google AI Plus, Pro, Ultra
Runs on: Gemini app for macOS
Control: pause, stop, approve actions
Claude computer use
Status: beta, off by default
Plans: Team and Enterprise
Runs on: Claude Desktop, macOS 14+ and Windows
Control: admin app block/allow lists, per-app approval
Apple Mac agent
Status: reported, early stage
Plans: not announced
Timing: as soon as late 2027, per Bloomberg
Related: macOS 27.3 warnings for third-party agents
ChatGPT dictation no longer stops at 10 minutes
The short version: Live Dictation now shows your words as you speak and keeps recording past the old 10-minute limit, on every ChatGPT plan.
OpenAI posted the change in its ChatGPT release notes on November 18, according to iThinkDiff. Dictation used to cut off at 10 minutes and only turned speech into text after you stopped talking. Now text appears live, and you can tap or click the waveform to show or hide it while you talk, then edit before you send. The update is rolling out worldwide on iOS, Android, the web and the desktop app, so it may take a few days to reach your account.
This is separate from voice mode, which talks back. Dictation simply types for you, which makes it useful for long briefs, meeting recaps and first drafts you want to clean up in the same chat.
Your move: if the waveform still stops at 10 minutes, update the app and try again later; the rollout is gradual.
How to turn off AI in Canva: the new setting for Free and Pro
The short version: Canva is rolling out a personal setting that turns off its generative AI features, but the step-by-step instructions are not live yet.
Canva's help center now has a "Turn off AI features" section for Canva Free and Pro accounts. AI features stay on by default, the switch is rolling out gradually, and Canva says the steps are "coming soon." It covers only AI features that already have a permission setting, and some low-risk tools such as background remover stay available either way. The choice does not carry over if your account moves between plan types.
A second, separate option already works: "Show less AI-generated content" filters AI-made templates and images out of search and recommendations. Go to your profile icon, then Your account, then AI personalization, and turn off AI-generated content. It does not remove AI from designs you already made. On Teams, Business, Enterprise and Education plans, admins control access through Canva's "Change access to Magic and AI features" page instead.
Your move: Free and Pro users who want less AI in search can switch the filter on today and check the help page again for the full off switch.
Meta Muse privacy report: what staff saw, and the $300 Muse Charm
The short version: Meta employees testing the Muse agent saw requests, orders and bank balances that were not theirs, according to Reuters.
A Reuters investigation published November 18, summarized by TechCrunch the next day, quotes internal posts in which employees said Muse showed them "childcare requests they never made" and "bank account balances that didn't belong to them." One employee said a weather chat came back with an added note listing bank balances. Another said Muse reminded her about a Signal message from her sister even though she had not given it access to other apps. Meta told TechCrunch that employee testing exists to surface feedback and that staff use different versions of Muse than the public. Reuters did not report confirmed exposure for public users.
Muse launched in September and has passed 10 million downloads. Separately, The Information reports Meta is aiming for a $300 price on Muse Charm, its keychain-sized Muse device, possibly with up to a year of free 5G data. Meta has only said it is targeting the holidays. We covered Muse's paid business calling feature on November 14.
Your move: open Muse's settings and review which apps and accounts it can read, and disconnect anything it does not need.
Did Claude solve the Riemann hypothesis? The claim is not confirmed
The short version: An Anthropic mathematician says he used Claude to produce a proof, but no outside expert or institution has verified it.
On Friday, November 20, mathematician Levent Alpöge, who works at Anthropic, said he had proved the Riemann hypothesis with Claude's help, Gizmodo reported. He says Claude "far outpaced" him near the finish and that the proof is written in Lean, a language that lets software check each step. As of Gizmodo's report, the result had not been confirmed by the Clay Mathematics Institute, by Anthropic or by any other credible third party, and no independent mathematicians had weighed in.
The hypothesis, first posed in 1859, is one of Clay's seven Millennium Prize Problems with a $1 million award. A Lean check makes errors in the logic much less likely, but experts still need to confirm that the formal statement matches the real conjecture. Alpöge said the timing next to Anthropic's IPO was "horrible" and rejected the cynical reading of it.
Your move: treat headlines saying "AI proved Riemann" as a claim, not a result, until mathematicians outside Anthropic finish their review.
Part 2 · SaaS and business
Anthropic stock (ANT) on day two: back near the $83 IPO price
The short version: ANT gave back most of its debut pop on Friday, trading around $86 at midday.
This is an update to our November 20 roundup of the ANT debut. Fortune reported at 11:44 a.m. ET on Friday, November 20, that shares were trading around $86, down about 11% on the session and only $3 above the $83 offering price. On Thursday the stock had touched $115.10 before settling at $97.17. The offering raised about $81.3 billion, priced above the original $70 to $78 range. We could not confirm Friday's closing price from a primary source at the time of writing.
For customers, nothing about Claude's plans or prices changed with the listing. For investors, the swing shows how quickly the market is repricing large AI names this month.
ANT share price, IPO to Friday midday (US$)
Salesforce layoffs, November 2026: what is confirmed so far
The short version: Salesforce has reportedly started another round of job cuts and capped employees' AI token spending, but the company has not confirmed numbers.
Business Insider reported the new cuts this week, based on internal messages and a manager's LinkedIn post, according to StockTwits and 24/7 Wall St. How many people are affected is unknown, and Salesforce has not publicly confirmed the round. StockTwits also reported that Salesforce introduced monthly AI token spending limits for employees after some staff ran up thousands of dollars in tokens last quarter.
The stock has taken the news hard. 24/7 Wall St. put Salesforce's last close at $183.31 on November 20, down 16.7% in a week and 27.7% this year, with the planned $5.6 billion Braze purchase adding to investor worries about cost. The next checkpoint is earnings on December 2 after the market closes, where the company has guided to $11.42 billion to $11.5 billion in revenue.
Your move: customers have no announced price or product changes to act on. If you are reviewing CRM contracts anyway, compare renewal quotes with our monday CRM vs HubSpot vs Pipedrive cost breakdown and the tools in our sales outreach hub.
Inception's Mercury API shuts down December 28: what developers must do
The short version: Inception is winding down its Mercury API and dashboard, and all data goes with them after December 28.
Inception's wind-down notice says the Mercury API and platform dashboard stay up through December 28, 2026. After that, requests stop being served and customers lose access to the platform and its data, so the company tells developers to export records and move integrations before the cutoff. The notice does not give a reason.
The likely context: The Information reports Nvidia is nearing a deal to buy Inception, and a TheFly summary put the price at no less than $500 million. Neither company has announced anything. Inception's diffusion-based Mercury models generate many tokens in parallel; the company says Mercury 3 reaches 1,068 tokens per second on Nvidia GPUs. There is no confirmed plan to release Mercury's weights.
Your move: if any feature in your product calls Mercury, pick a replacement model now and budget for the price difference. Our SaaS price change tracker lists other recent shutdowns and price moves to check while you migrate.
Weekend checklist from today's news
- NowClaude Team/Enterprise owners: decide whether to enable computer use, and set an app allow list first.
- NowGemini AI Plus, Pro or Ultra on a Mac: test computer use on a harmless file task with approvals on.
- NowCanva Free/Pro: turn off AI-generated content in search under Your account, then AI personalization.
- NowMuse users: review connected apps and remove access it does not need.
- Dec 2Salesforce reports earnings after the close; watch for layoff and pricing details.
- Dec 28Inception's Mercury API and dashboard go dark; export data before then.
Your questions on Gemini, Claude, Canva and the Riemann claim
Is Gemini computer use on Mac free?
No. Google limits it to Google AI Plus, Pro and Ultra subscribers using the Gemini app for Mac, and only in English in the US for now. There is no separate charge on top of those plans, but free Gemini accounts cannot use it, and Google has not said when other countries will get it.
How do I turn on Claude computer use for my team?
An Owner or Primary Owner opens Organization settings, then Capabilities, confirms "Chat and Cowork unified" is on and switches on Computer use under Claude Desktop. Each person then enables it on their own computer and approves apps one by one. Admins can block apps or allow only a set list.
Can I turn off all AI features in Canva?
Not completely yet. Canva Free and Pro users are getting a "Turn off AI features" setting gradually, and it only covers features that already have a permission control. Low-risk tools such as background remover stay on. Team and Enterprise admins manage AI access through a separate admin setting.
Has the Riemann hypothesis been proven by AI?
Not officially. An Anthropic mathematician says he used Claude to write a Lean proof, but as of November 20 the Clay Mathematics Institute, Anthropic and outside experts had not confirmed it. Until independent mathematicians check that the formal statement is right, it remains a claim.
Sources: TechCrunch, Gemini on Mac (Nov 19) · TechCrunch, Apple agent report (Nov 19) · Claude release notes (Nov 20) · Claude help center, computer use setup (Nov 20) · iThinkDiff, ChatGPT Live Dictation (Nov 18) · Canva help center, manage AI features (checked Nov 21) · TechCrunch on Reuters' Muse report (Nov 19) · Gizmodo, Muse Charm price (Nov 19) · Gizmodo, Riemann claim (Nov 20) · Fortune, ANT day two (Nov 20) · Anthropic IPO pricing (Nov 18) · StockTwits, Salesforce cuts (Nov 17) · 24/7 Wall St., Salesforce (Nov 20) · The Neuron, Nvidia and Inception (Nov 20) · Inception wind-down notice