### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) and this hasn't been reported yet - [x] This is a single bug report (please file separate reports for different bugs) - [x] I am using the latest version of Claude Code ### What's Wrong? **What's Wrong?** Claude Code appears to retain inconsistent account identity and authentication state after switching between two Claude Max accounts.

[BUG] Account identity and usage quota mismatch after switching Claude Max accounts

Status Open

Reported on v2.1.299

Maintainer reply None cached

Activity 0 comments · opened Oct 14, 2026

Preflight Checklist

[x] I have searched existing issues and this hasn't been reported yet
[x] This is a single bug report (please file separate reports for different bugs)
[x] I am using the latest version of Claude Code

What's Wrong?

What's Wrong?

Claude Code appears to retain inconsistent account identity and authentication state after switching between two Claude Max accounts.

The default profile (~/.claude) stores the identity and organization metadata of Account A (Max 20x), while live usage measurements from that profile match Account B (Max 5x).

At approximately 18:10 on October 14, 2026, minimal requests from both profiles were reflected in the same usage bucket (38% session / 48% weekly), with identical reset times.

Later, the affected default-profile session reported 44% session usage and 49% weekly usage, with a weekly reset on October 17 at 14:00 Europe/Berlin. An independently authenticated, isolated Account A profile instead reported a weekly reset on October 20 at 18:00.

Remote Control also reported an active session in the affected profile, but the session was unavailable in the expected account's mobile app. A causal connection has not been established.

This suggests that locally cached account metadata and the credentials actually used for requests can become inconsistent. The exact token identity and root cause have not been conclusively established.

What Should Happen?

What Should Happen?

Claude Code should consistently associate the active credentials, account identity, organization, subscription plan, and usage quota with the same authenticated account.

Switching accounts should not leave a running session or background credential-refresh process using credentials from a different account while displaying stale account metadata.

If the active credentials do not match the stored account identity, Claude Code should detect the mismatch and request reauthentication instead of silently continuing.

Error Messages/Logs

**Error Messages/Logs**

No directly attributable authentication error was captured for the quota mismatch.

The affected Remote Control session reported itself as active, but the mobile application displayed:

"This session is unavailable."

A separate service reported:

403 — login carries no Claude Design access.

Two historical 401 errors ("OAuth access token has expired") were found from September 16, but their relationship to this issue is unknown.

No tokens, cookies, credentials, or unredacted account identifiers are included.

Steps to Reproduce

Steps to Reproduce

The original account mismatch was observed after account switching in an existing Windows Claude Code environment. The precise sequence that first introduced the mismatch is not yet known.

Use two independently authenticated Claude Max accounts with separate Claude Code profiles.
Start a long-running Claude Code session using the default profile.
Switch between the accounts using the normal Claude Code authentication flow while other sessions or background processes remain active.
Compare the stored non-secret account metadata with the usage limits reported by /usage and the live status line.
Run a minimal request from each account profile and compare the usage buckets and reset timestamps.
Observe whether the default profile displays Account A's identity while its live usage tracks Account B's quota.

Observed on October 14, 2026: the default profile and Account B produced matching live usage measurements, while a newly isolated Account A profile showed a different weekly reset.

This is an evidence-based reproduction outline, not yet a reliably repeatable minimal reproduction. No credential manipulation was performed.

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.299 (Claude Code)

Platform

Other

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

Additional Information

Environment:

Windows
Claude Code 2.1.299
Claude Max 20x and Max 5x accounts
Multiple local Claude Code profiles
Long-running sessions and background token refresh activity

An isolated Remote Control authorization test was also performed with two freshly authenticated profiles.

Results:

Account A could access its own test session.
Account B could access its own test session.
Account A was denied access to Account B's session.
Account B was denied access to Account A's session.

No unauthorized cross-account access was demonstrated.

Local session transcripts contain historical Remote Control registrations associated with different, internally consistent account/organization pairs.

A background credential-refresh process retaining an older token is one possible explanation, but this has not been verified.

Redacted evidence can be provided upon request. Please advise if additional diagnostic logs would help reproduce the issue.

Both accounts use Claude Max subscriptions through Claude Code, not API keys or usage-based Anthropic API billing.

Additional evidence — two simultaneous sessions on the same Windows machine

I captured two Claude Code /usage screens on October 14, 2026, from separate sessions running simultaneously on NODE-01.

Existing long-running session (default profile):

Current session usage: 46%
Weekly usage: 49%
Weekly reset: October 17, 14:00 (Europe/Berlin)
Session API-equivalent cost: $23.27

Fresh isolated session (Account A, Max 20x):

Current session usage: 0%
Weekly usage: 0%
Weekly reset: October 20, approximately 18:00 (Europe/Berlin)
Session API-equivalent cost: $0.0974
Successfully authenticated and processed a real model request

Both sessions ran on the same Windows computer.

The default profile's locally stored account metadata identifies Account A (Max 20x), but its usage values closely match the independently authenticated Account B (Max 5x) profile.

This provides additional evidence of an account identity/usage quota mismatch. The exact active token identity in the long-running session has not been independently verified.

Screenshots of both /usage views are attached.

View original on GitHub ↗