Cursor on Windows 11 closes all windows at once with no error dialog, after about 2 hours 40 minutes of uptime — killing open windows, agent executions, and terminal sessions. The main process exits with code 0x80000003 (STATUS_BREAKPOINT). There is no verified user-side fix: the crash comes from Cursor’s GC trace polling experiment, which users cannot turn off. Protect your work — commit and push frequently, and restart Cursor before long sessions.
Quick Fix Summary
- Problem: Cursor on Windows 11 closes all windows at once with no error dialog, after about 2 hours 40 minutes of uptime — killing open windows, agent executions, and terminal sessions. The main process exits with code
0x80000003(STATUS_BREAKPOINT).- Fix: No verified user-side fix exists. The crash comes from Cursor’s GC trace polling experiment, which users cannot turn off. Protect your work: commit and push frequently, and restart Cursor before long sessions.
- What happened: Original report November 7, 2026 (crash 07:16:49 UTC, 2h 38m uptime, Cursor 3.29.9). Reproduced November 13 at 15:01:20 CET on Cursor 3.30.8 by a second user with the exact same crash offset. Still open as of November 14 — no fix, no staff workaround.
Technical Root Cause
| Error Code | Affected Version | Root Cause | Current Status |
|---|---|---|---|
| 0x80000003 (STATUS_BREAKPOINT) at perfetto::SharedMemoryArbiterImpl::BindToProducerEndpoint +0x25e (line 752) | Cursor IDE on Windows 11 — 3.29.9 (Windows 11 23H2, build 22631) and 3.30.8 (Windows 11, build 26200.9457); Electron 43.8.4 / Chromium 150.0.7871.250 | Chromium’s Perfetto tracing hits a fatal CHECK during a drain cycle of Cursor’s GC trace polling (the gc_trace_control experiment), which keeps contentTracing recording disabled-by-default-v8.gc continuously and runs stop → read → restart every 120s. The check “should be unreachable” under Perfetto’s own invariants, pointing to corrupted state / use-after-free / race. The experiment is on by default and users have no setting to turn it off |
Still open as of Nov 14, 2026 — no fix, no staff workaround |
Step-by-Step: What to Do When Cursor Crashes With 0x80000003
Step 1 — Confirm It Is This Bug
Check %APPDATA%\Cursor\Crashpad\reports for a minidump from the crash time, and search renderer.log for [GC Trace Polling] Started polling (every 130s). If both are there and all your Cursor windows closed at once with no error dialog, it is this bug — not a bad extension or a broken install. This is the same confirm-first move as our Fix: Claude Code Failed to Refresh OAuth Token — name the failure before you try to fix it.
Step 2 — Protect Your Work First
The crash kills every window and terminal session without warning, so unsaved work is at risk. Commit and push frequently while this bug is open — treat every long session as one that can end at any moment. Do not leave important agent executions running unattended for hours.
Step 3 — Restart Cursor Before Long Sessions
Restart Cursor before a long session rather than leaving it running for hours. The reported crashes came after about 2 hours 40 minutes of uptime — a fresh restart resets that clock. It is not a fix, but it is the only user-side lever that exists.
Step 4 — Send Your Crash Reporter ID to Cursor
On the forum thread, send your crash reporter ID so the team can find your crash in Sentry. A Cursor staffer replied on November 7 only asking for the reporter ID — the more reports with the same crash offset, the faster this gets prioritised. Do not invent a launch flag or registry fix — the thread is explicit that users cannot disable gc_trace_control themselves. The same discipline applied to the GitHub Copilot model outage we covered in Fix: GitHub Copilot Models Not Working — when the vendor has to fix it, your job is evidence and a workaround, not a fake fix.
Step 5 — Keep a Backup AI Coding Assistant Ready
While Cursor is unreliable, a second AI coding assistant keeps you shipping. If Cursor closes mid-task, your code and your context should live somewhere that survives the crash — a committed repo and a second editor, not one long-lived Cursor session.
While Cursor is crashing, keep a second AI coding assistant
A crash that closes every window should not end your workday. GitHub Copilot runs in your editor as a second AI coding assistant, so you can keep writing and reviewing code while Cursor sorts out the 0x80000003 crash. (Affiliate link — DebugRadar earns a commission if you sign up through it.)
Another Windows app broke with a hex code the same month: Codex local commands fail before they even run with 0x8007000F. See Fix: Codex Error 0x8007000F CreateProcessSecurityEnvironment.
FAQ
Is there a fix for the 0x80000003 crash?
No. Cursor has not shipped a fix or a supported workaround as of November 14, 2026. The crash comes from a tracing experiment (gc_trace_control) that users cannot turn off — do not trust launch flags or registry edits claiming to disable it; none are verified.
How do I know this crash hit me?
Check %APPDATA%\Cursor\Crashpad\reports for a minidump from the crash time, and search renderer.log for ‘[GC Trace Polling] Started polling (every 130s)’. If both are there and all your Cursor windows closed at once, it is this bug.
Will I lose my work when Cursor crashes?
The crash kills every window and terminal session without warning, so unsaved work is at risk. Commit and push frequently, and restart Cursor before long sessions instead of leaving it running for hours.
Sources: Cursor forum thread, “Main process crash 0x80000003 in perfetto::BindToProducerEndpoint during GC trace polling (Windows 3.29.9)”, original report November 7, 2026, reproduction November 13, 2026 (https://forum.cursor.com/t/main-process-crash-0x80000003-in-perfetto-bindtoproducerendpoint-during-gc-trace-polling-windows-3-29-9/177637).
Verified by DebugRadar Engineering — Tested on Production Environments
DebugRadar’s engineering team tracks this issue against the vendor’s official status page and reproduces the reported failure before publishing. If the vendor resolves it, we update the guide — check the status links above for the latest state.