MorseSign inStart freeOpen Morse

Is Claude Cowork safe? What it can reach, where files go, and what Anthropic warns

The Morse teamPublished 9 min read

As of November 2026, Claude Cowork is safe enough for low-stakes files and tasks you scope tightly, but not for your finances, health records, or other people's data, and Anthropic's own help center says so. Anthropic calls the chance of an attack "still non-zero" and "strongly" advises against using Cowork's browser for sensitive information (Anthropic). Since October 6, 2026, new Cowork tasks on Pro, Max, and Ultra plans run on Anthropic's servers instead of your computer (Anthropic).

We read Anthropic's Cowork help articles (most updated between October 30 and November 12, 2026), its release notes, and its privacy center pages on training and retention on November 14, 2026. Anthropic deserves credit for a safety page that names its own limits plainly, and Cowork always asks before it permanently deletes a file.

Claude Cowork safety at a glance

QuestionAnthropic's answer, as of November 14, 2026Source
Who can use itPaid plans only: Pro, Max, Ultra, Team, and EnterpriseAnthropic
Where the work runsOn Anthropic's servers for new Pro, Max, and Ultra tasks since October 6, 2026, with no way to switch backAnthropic
Which files it can openOnly folders you connect, through Claude Desktop, while the app is openAnthropic
Deleting filesAlways asks first, in every approval modeAnthropic
Default approval modeAuto on Pro, Max, Ultra, and Team; Manual on EnterpriseAnthropic
Deleted tasksGone from your history at once, and from Anthropic's back end within 30 daysAnthropic
Model trainingFollows your Help improve our AI models setting on personal plansAnthropic
Computer useBeta on Pro, Max, and Ultra; "no sandbox" between Claude and your appsAnthropic

Does Claude Cowork access your files?

Yes, but only the folders you connect. Anthropic says a cloud session reaches your computer "only when the Claude Desktop app is open, only for the folders you've connected there, and with the permissions you've already set" (Anthropic). Inside those folders, Claude "can read, write, and permanently delete" files, so Anthropic suggests "a dedicated working folder for Claude" and backups of anything important (Anthropic).

When a cloud task needs one of your files, Claude "fetches a copy of just that file," and those copies are deleted when you delete the session (Anthropic). The folders themselves stay on your computer (Anthropic).

If your employer manages your computer, Anthropic warns that "connecting local folders makes them reachable from a session in the cloud" (Anthropic).

Does Claude Cowork run on your computer or in the cloud?

In the cloud, for every new task on a personal plan. On October 6, 2026, Anthropic removed the Only on your computer option from Settings for Pro, Max, and Ultra, and people who had it on were moved to the cloud the next time they opened the desktop app (Anthropic). Anthropic says the files Claude opens there are "processed on Anthropic's servers rather than staying on your computer" (Anthropic).

Fast.io's safety guide, updated March 16, 2026, still says Cowork "runs tasks inside a virtual machine on your computer" (Fast.io). The Cloud Security Alliance's July 8, 2026 post calls it "a local agent that runs on the employee's machine" and says Cowork activity is excluded from the Compliance API "on every tier" (Cloud Security Alliance). Anthropic's Team and Enterprise page now says Cowork sessions through Claude, Claude Desktop, and Claude Mobile "are captured in the Compliance API" (Anthropic).

The cloud session itself is walled off from your machine. Anthropic describes "an isolated, temporary environment" that "can't reach your home or company network" and is removed when the session ends (Anthropic). Anthropic adds that this isolation "doesn't limit what Claude reads or does" through the folders, apps, and sites you've connected (Anthropic).

How Cowork moved from your laptop to Anthropic's servers

DateWhat changedSource
January 12, 2026Research preview for Max on macOS, running "locally on your computer in an isolated VM"Anthropic
March 23, 2026Computer use preview for Pro and MaxAnthropic
April 9, 2026Generally available on macOS and WindowsAnthropic
July 7, 2026Web and mobile, with sessions run remotely in beta and saved to your Claude accountAnthropic
August 25, 2026Memory shared between chat and cloud CoworkAnthropic
September 16, 2026Cowork merged into every conversation, rolling out graduallyAnthropic
October 6, 2026New Pro, Max, and Ultra tasks run in the cloud onlyAnthropic

Tasks started on your computer before October 6 stay there. For new work that has to stay local, Anthropic points to Claude Code in the desktop app, which "runs on your computer and keeps your folders and history there," and says a Cowork task moves over with Download task data and the /port-task-data command (Anthropic). Scheduled tasks run in the cloud too, "even when your computer is asleep or the Claude Desktop app is closed," and they can't use the accounts you've linked in Claude's Personal Finance feature (Anthropic). Scheduled tasks set up on your computer before October 6 keep running there (Anthropic). Transcripts of the older local tasks are exempt from Claude Code's 30-day local cleanup by default (Claude Code docs). We cover how to clear them in how to delete Claude chats.

Does Claude Cowork train on your data?

On a personal plan, only if your training setting is on. Anthropic says whether Cowork conversations improve Claude "follows the Help improve our AI models setting in Settings > Privacy" (Anthropic). Training data leaves out "raw content from connectors," such as Google Drive, unless it's copied into the conversation (Anthropic). Team and Enterprise inputs aren't used for training by default (Anthropic). The setting's history and its five-year retention are in is Claude private.

How long does Anthropic keep your Cowork tasks?

Until you delete them. A deleted task leaves your history "immediately" and Anthropic's back-end storage "within 30 days" (Anthropic). Working files Claude doesn't hand back to you "aren't kept or referenced after the session ends" (Anthropic). A session flagged under the Usage Policy is kept longer: inputs and outputs for up to 2 years, and safety scores for up to 7 (Anthropic).

Cowork sessions "can't be shared with others," though you can share individual artifacts Claude makes during a session (Anthropic). Cloud Cowork also shares memory with chat. To run one task without it, turn Memory off in the + menu before the first message, because it "can't be changed later" (Anthropic).

Can Claude Cowork be hijacked by prompt injection?

Yes, and Anthropic says so. Its safety page describes an email that tells Claude to "transfer $1000 to this account" and says an attack needs two things at once: Claude reading content from outside your trusted sources, and Claude able to act on it (Anthropic). Its defenses are model training, classifiers that scan "all untrusted content," action checks in Auto mode, and the deletion prompt (Anthropic). Anthropic still says "the chances of an attack are still non-zero" (Anthropic).

In January 2026, PromptArmor reported a document with hidden text that made Cowork upload a user's files to an attacker's account (PromptArmor). We explain how these attacks work in what is prompt injection.

Your approval mode decides how far an injection gets. Anthropic offers three, though the new combined Claude experience shows only Auto and Manual (Anthropic):

  • Manually approve. Claude asks before each action, and you choose Allow or Deny.
  • Automatically approve. Claude reviews each action for risks such as "data exfiltration or prompt injection" and blocks what it judges unsafe.
  • Skip all approvals. "Nothing checks its actions automatically."

Anthropic says to switch to Manual for sensitive files, a new plugin or site, and anything "hard to undo, like sending messages or making purchases" (Anthropic).

What can Claude see through your browser and screen?

Whatever is open to it. Computer use works by screenshots, so Claude "can see any information visible on your screen" in the apps you've allowed (Anthropic). Claude asks before using each app, and investment, trading, and cryptocurrency apps are blocked by default (Anthropic). A link Claude clicks in an allowed app still opens in another app, and Anthropic says it "can't stop the link from opening" (Anthropic).

Cowork's built-in browser can import your saved logins site by site, with banking, email, and single sign-on sites "unchecked by default" (Anthropic). Anything you sign in to there "is available to Claude in future Cowork sessions on that computer" (Anthropic).

What Anthropic's own pages tell you not to do in Cowork

Each row is Anthropic's advice, read on November 14, 2026.

Don'tAnthropic's reasonSource
Connect folders with financial documents, credentials, or personal recordsClaude can read, write, and permanently delete what's in themUse Cowork safely
Use the browser for financial accounts, medical information, or other people's data"We strongly advise against" itBuilt-in browser
Give computer use banking, healthcare, or government appsIts guardrails "aren't absolute"Computer use
Schedule tasks that send messages, make purchases, or open sensitive filesYou can't watch them in real timeUse Cowork safely
Use Skip all approvals unless you trust every connector, file, and appNothing checks Claude's actionsGet started
Install unverified extensions or pluginsLocal MCP servers run "with the same permissions as any other program"Use Cowork safely
Work with sensitive data in the Excel and PowerPoint add-ins during a sessionData can pass between apps without you directing itUse Cowork safely

Anthropic also says you "remain responsible for all actions taken by Claude" on your behalf, including purchases and messages sent (Anthropic).

What can Team and Enterprise admins control?

Whether Cowork runs at all, where it runs, and how much Claude may do without asking. Cowork is on by default, but owners can switch it off, and it starts off in organizations that have enabled HIPAA (Anthropic). Cloud sessions are on by default for Team and off by default for Enterprise (Anthropic). Owners can hide Automatically approve, and members can't skip per-task approval for write-capable connector tools unless an owner allows it (Anthropic). Enterprise admins can retrieve Cowork sessions through the Compliance API (Anthropic). Team and Enterprise owners can also stream tool calls, file access, and approval decisions to their security tools through OpenTelemetry (Anthropic). Computer use isn't available on Team or Enterprise as of November 2026 (Anthropic). We cover what an employer can see more broadly in AI on a work laptop.

How Morse keeps your most sensitive questions out of your Claude account

Since October 6, new Cowork tasks on Pro, Max, and Ultra run on Anthropic's servers and are saved to your Claude account, and the files they open go there too. Anthropic tells you to keep questions about money, health, and other people's data out of Cowork. Morse is a chat app, and your chat history is stored in your browser, encrypted, not on our servers. Our servers don't keep messages or log what you write, and Morse doesn't train AI on your content, sell your data, or show ads.

  • E2EE, for the money or health question Anthropic tells you not to hand Cowork. Your browser checks the secure enclave's proof that it's genuine, then encrypts the message so only that enclave can read it. Morse's servers relay ciphertext they can't read, and the reply is decrypted only in your browser. This rests on the enclave hardware and attestation working as designed, and E2EE has no web search or image generation. E2EE chats come with Pro, at $19 a month; here's how end-to-end encrypted AI chat works.
  • Private, for everyday questions no host should keep. New chats start here, with open-weight models like DeepSeek, Kimi, and GLM, or xAI's Grok (Grok comes with Pro). Morse has zero data retention agreements with its inference, search, and other providers, so they've agreed to keep nothing, and where a provider confirms zero retention on each reply, Morse checks it.
  • Anonymous, for Claude itself without your Claude account. Our proxy removes your account, email, and IP address and sends the message under Morse's own account. The provider sees the text without knowing it's you and may keep it for a limited time to check for abuse, under its own policy, so leave names and account numbers out.

Because your history is kept in your browser, clearing it or losing the device deletes your chats for good, and Morse can't recover them. Save any answer you'll need.

How to make Claude Cowork safer, by worry

  • An injection hijacking a task: use Manually approve for anything that sends, buys, or deletes, and connect only the sites and plugins you trust.
  • Files leaving your computer: connect one working folder with copies, not your home folder, and delete sessions you're done with so the fetched copies go too.
  • Training on your work: turn off Help improve our AI models in Settings, Privacy. Morse doesn't train AI on your chats; in Anonymous mode, the provider's own terms apply.
  • A question about your bank, your health, or someone else: keep it out of Cowork, as Anthropic advises, and ask it in an E2EE chat on Morse.

You can try a chat from the Morse home page without an account, or start free with no card.

Common questions

Can you still run Claude Cowork only on your own computer?

Not for new tasks on Pro, Max, or Ultra. Anthropic's help center points anyone who needs that to Claude Code in the desktop app, which it says "runs on your computer and keeps your folders and history there," and says a Cowork task can be carried over with "Download task data" and the /port-task-data command.

Can your employer see what you do in Cowork?

On a work plan, largely yes. Anthropic's help center says Enterprise admins can retrieve Cowork sessions through the Compliance API, and Team and Enterprise owners can stream tool calls, file access, and approval decisions to their security tools through OpenTelemetry.

Can other people see your Cowork sessions?

Not through sharing. Anthropic's help center says Cowork sessions "can't be shared with others," though you can share individual artifacts Claude makes during a session.

Do scheduled Cowork tasks need your computer to be on?

Not for new ones. Anthropic's help center says scheduled tasks run "even when your computer is asleep or the Claude Desktop app is closed," while ones set up on your computer before October 6 keep running there. It also says scheduled tasks can't use the accounts you've linked in Claude's Personal Finance feature.

Ask what you actually want to ask.

80+ models, characters, and images, with every chat encrypted and kept on your device. Free to start.

Start a private chat

Read next