Daily morning Authentication error after overnight Cursor session (Windows + Enterprise)

Where does the bug appear (feature/product)?

Cursor IDE (Agent / Chat)

Describe the Bug

Every morning when I start using Cursor, Agent chat shows:

Authentication error
If you are logged in, try logging out and back in.

UI still looks logged in. After Sign Out + Sign In again, it works for the rest of the day. The problem comes back the next morning.

This is reproducible as an overnight / long-lived session issue, not a bad password.

Request ID

83a21c60-0790-4200-aa3d-64c3c1ddb77b

Steps to Reproduce

  1. Keep Cursor running overnight (do not fully quit).
  2. Next morning, open Agent chat and send any message.
  3. Observe Authentication error / ERROR_NOT_LOGGED_IN.
  4. Sign out + sign back in → works until the next overnight period.

Operating System

Windows 11 (Windows_NT x64 10.0.22631)

Version Information

  • Cursor: 3.14.7 (stable)
  • Commit: a758f2241ca99fecf380180b6cbdbbce0f1f42c0
  • Electron / layout: unifiedAgent + glass also used

Account

  • Enterprise membership
  • Auth via Auth0 (signupType=Auth_0)
  • Corporate network / Tencent environment

Local diagnostics already completed

We ruled out the common local causes suggested in similar threads:

Check Result
Windows clock sync OK — NTP synced to time.tencent.com, timezone Asia/Shanghai (+08:00), clock is correct
System / WinHTTP proxy Disabled / direct (no system proxy)
Custom Model Base URL Not set
HTTP env proxies (HTTP_PROXY etc.) Empty
Network reachability api2.cursor.sh returns 200
Tokens present after re-login cursorAuth/accessToken + refreshToken both present
Token JWT exp after re-login ~2026-10-04 (not short-lived expiry explaining morning failure)

Strong evidence from local logs

Cursor process was left running from 2026-08-04 11:21.

Starting around midnight, logs show repeated:

ConnectError: [unauthenticated]
error: ERROR_NOT_LOGGED_IN
title: Authentication error
detail: If you are logged in, try logging out and back in.

In one overnight session we counted ~1538 ERROR_NOT_LOGGED_IN log hits from about 00:05 until around 10:04 (when usage / re-login resumed).

Also observed:

  • Transport uses HTTP/2 to Cursor APIs
  • Logs show tls.hasCustomCA: true (corporate CA trust store present)
  • At window restore time: Failed to refresh membership ... {"error":{"name":"Canceled"}}
  • Auth race log: auth_ready_race_prevented with large waitedMs earlier in the session

This matches the known client-side auth-refresh / token-attachment breakage discussed in:

Mitigations applied locally

  1. Confirmed clock sync is healthy.
  2. Enabled cursor.general.disableHttp2: true (HTTP/1.1 fallback) as suggested for auth-refresh instability.
  3. Temporary workaround remains: full Sign Out → fully quit Cursor → Sign In again each morning.

Expected Behavior

Long-lived overnight sessions should successfully refresh auth and keep Agent usable the next morning without manual re-login.

Does this stop you from using Cursor?

Yes — every morning until re-login. Blocks Agent until manually fixed.

Ask for Cursor team

Please check backend/client traces for Request ID 83a21c60-0790-4200-aa3d-64c3c1ddb77b, and whether overnight token refresh is failing client-side before the request is sent (as described in the ERROR_NOT_LOGGED_IN investigation). Happy to capture Network tab auth/refresh request details on the next failure if needed.

Hey there,

What you’re describing isn’t caused by anything in your setup (clock, proxy, HTTP/2, or token expiry). On the current stable build, a long-lived overnight session can leave Agent requests going out without an auth header, so you get Authentication error / ERROR_NOT_LOGGED_IN even though the UI still looks signed in. Same class as Workspace agents fail with missing authorization header on macOS 3.14.7. Turning off HTTP/2 won’t help here.

Until a fixed build reaches the normal update channel, try Developer: Reload Window (Ctrl+Shift+P) on the affected window first - lighter than a full sign-out. If that doesn’t stick, your Sign Out → fully quit → Sign In path is the reliable workaround for the day.

We’ve let the team know and this is an issue we’re tracking. I’ll post here when there’s an update you can install.