On October 6, 2026, Cursor's changelog announced that you can now message the agents running on your laptop from the Cursor iOS app, and added: "Remote Control is on by default for everyone except Enterprise organizations." A month earlier, Claude Code users had started filing GitHub issues saying the same feature in their tool switched itself on. If you build with either, the agent on your machine can probably be steered from your phone right now. Here's how to check, as of 2026-10-12, and how to turn it off.
TL;DR
Remote Control lets a phone or browser send instructions to the AI agent on your laptop, and that agent has your .env, database URLs and deploy tokens. Cursor's changelog says it's on by default (except Enterprise), though Cursor's own docs disagree on several details. Claude Code documents it as opt-in, but users report it turning on anyway. In Claude Code, set remoteControlAtStartup to false explicitly, or disableRemoteControl to true. In Cursor, check Settings > Agents in the Agents Window.
Why this matters for an app you shipped with an agent
In both tools, the commands still run on your laptop. Claude Code says execution and filesystem access "stay on your machine". Cursor's docs say "The agent loop runs in Cursor's cloud" while terminal commands, file edits and tests run on your computer. Either way, the tools touch the laptop that holds your production DATABASE_URL, your Stripe secret key and your deploy CLI login, and now something other than your keyboard can send them work.
This isn't a remote takeover. Both tools tie sessions to your own account, and Cursor's changelog says pairing a phone needs approval in the desktop app. What changes is who has to be careful:
- Your phone and browser become a way into your laptop's shell. If your phone is unlocked, lost, or signed in somewhere you share, the agent on your laptop takes instructions from it.
- The conversation leaves your machine. Claude Code's docs say that while Remote Control is connected, "the session transcript, including your messages, Claude's responses, and tool activity, is stored on Anthropic servers." If the agent ran
cat .envin that session, treat the output as part of what's stored. Cursor's docs say Remote Control "isn't available when your privacy settings disable cloud data storage", so Cursor needs cloud storage on for it to work.
If you never use your phone to drive an agent, there's no reason to leave either one on.
Claude Code: check it and turn it off
Look at the setting. Run /config inside Claude Code and find Enable Remote Control for all sessions. The docs list three values: true, false, and default, which follows "your organization's admin default if one is set, otherwise Claude Code's current default." Pick false, not default, because default is the value that lets someone else decide. Two things still beat a user-level false: a true in your organization's managed settings, and passing --remote-control yourself for one session.
Pin it in your settings file. Add this to your user settings at ~/.claude/settings.json:
{
"remoteControlAtStartup": false
}
To turn Remote Control off everywhere, so Claude Code refuses claude remote-control, the --remote-control flag, auto-start and the in-session toggle, use disableRemoteControl instead:
{
"disableRemoteControl": true
}
Turn it off in the Desktop app and VS Code too. They have their own toggles. In the Desktop app it's Settings > Claude Code > Connect new sessions to Remote Control. In the VS Code extension it's Enable Remote Control for all sessions in the command menu's Settings section.
Check for sessions you didn't start. Remote Control sessions show up in your own Claude apps and at claude.ai/code. If you see sessions from your machine there that you never connected on purpose, that's your answer.
A repo can only turn it off, never on. Claude Code honors "remoteControlAtStartup": false in a project's .claude/settings.json but ignores true there, "so a checked-in file can't turn on Remote Control for everyone who opens the repository." If your project holds production credentials, commit the false.
What the bug reports say
Claude Code's docs say Remote Control "only activates when you explicitly run" claude remote-control, --remote-control or /remote-control, unless auto-connect is turned on. Some users say it didn't work that way for them.
- Issue #90341, opened August 28, 2026: "Remote Control auto-enables without user consent on new sessions." The reporter says they never ran any Remote Control command. Two other users confirmed it, one on September 6 on both macOS and Windows 11. That commenter, reading the CLI binary, reported that an explicit
remoteControlAtStartupsetting takes priority over a server-side default, and that setting"remoteControlAtStartup": falseand"autoUploadSessions": falsein~/.claude/settings.jsonstopped it. - Issue #92704, opened September 7, 2026: Remote Control "appears active on a new session despite being disabled" in the Desktop app.
As of 2026-10-12, both issues are open with no reply from Anthropic. We can't confirm the cause from the outside, and the binary analysis is a user's reading, not documentation. The safe conclusion doesn't depend on it, though: set the value to false explicitly rather than trusting the default.
Cursor: check it and turn it off
Cursor's own sources disagree about how this works, in several places, so check rather than assume.
| Changelog (Oct 6, 2026) | iOS docs | |
|---|---|---|
| Default | On for everyone except Enterprise | You turn it on under Settings > Agents |
| Where the agent runs | "doesn't move your agents anywhere" | "The agent loop runs in Cursor's cloud" |
| Cloud Agents | "doesn't require cloud agents to work" | Offered "for users with Cloud Agents access" |
| How you connect | Pair from the iOS app, approve on desktop | Run /remote-control to hand off a session |
| Admin control | Org settings > Security and identity > Remote control (Enterprise) | Cursor Dashboard > Cloud Agents > Self-Hosted (Teams and Enterprise) |
Neither page tells an individual user how to turn it off, and neither describes unpairing a phone. Here's what you can do with what's documented:
Open Settings > Agents in the Agents Window and find the Remote Control setting. The docs only describe turning it on there, so switching it off in the same place is the obvious move but not a documented one. They also note that "older clients won't show the Remote Control setting" and require Cursor 3.9.8 or later, so update if it isn't there.
Check the iOS app if you installed it. Per the changelog, computers on your account appear in the app automatically, and pairing needs your approval on the desktop. If you remember approving a request you didn't expect, treat that as the thing to undo. Cursor doesn't document how to unpair, so the desktop setting in step 1 is the control you can actually rely on.
On a Teams or Enterprise plan, ask your admin. The docs say "When it's off, members can't enable Remote Control." That's the cleanest control if several people's laptops hold production credentials.
Lock down the account, not just the toggle. Both features are tied to your Cursor or Claude account, so whoever controls that account controls the bridge. Turn on two-factor authentication for both. Claude Code also has a beta Trusted Devices setting that's off by default; on Pro and Max you turn on Require trusted devices yourself.
If you do want to use it
Remote Control is useful. Checking on a long agent run from your phone beats walking back to your desk. If you keep it on, keep the blast radius small:
- Don't run the agent from a directory where
.envholds production secrets. Use development credentials locally and keep production keys in your host's secret store. - Review what the agent does from your phone the way you would at your desk. A small screen makes it easy to approve a command you didn't read.
- Start Remote Control per session when you need it (
/remote-controlin Claude Code) instead of auto-connecting every session.
Remote Control doesn't leak anything by itself. It adds one more way for the agent to be told to do something with what it can reach.
Is Cursor Remote Control on by default?
Yes, for everyone except Enterprise organizations, per Cursor's October 6, 2026 changelog. Cursor's docs put the setting under Settings > Agents in the Agents Window, require Cursor 3.9.8 or later, and say the agent loop runs in Cursor's cloud while tools run on your computer.
How do I turn off Remote Control in Claude Code?
Run /config and set Enable Remote Control for all sessions to false, not default. Or add "remoteControlAtStartup": false to ~/.claude/settings.json. To block it completely, including claude remote-control and the --remote-control flag, add "disableRemoteControl": true. In the Desktop app, turn off Settings > Claude Code > Connect new sessions to Remote Control.
Why did Claude Code Remote Control turn on by itself?
Anthropic hasn't said. Claude Code's docs say Remote Control only starts when you run it or turn on auto-connect, but GitHub issues #90341 and #92704 report it starting on new sessions without either. One commenter reported that setting remoteControlAtStartup (and autoUploadSessions) to false explicitly stopped it.
Can someone else control my agent through Remote Control?
Not without your account. Both vendors tie sessions to your own account, and Cursor's changelog says pairing a phone needs approval in the desktop app. The risk is anyone who has your signed-in phone or browser, plus the session itself living partly in the vendor's cloud while it's connected.
Check what your agent already shipped
Scan your deployed app for exposed keys, open files and missing headers. No signup needed to see your first results.