Detailed Analysis
A user-reported bug affecting access to Claude Design — a product or feature route accessible via claude.ai or claude.com/design — has surfaced on the Anthropic subreddit, describing a persistent authentication redirect loop that prevents users from reaching the application. The reported behavior follows a consistent pattern: a user who is already authenticated on claude.ai attempts to navigate to the Design route, only to be caught in a cycle where a login redirect triggers a successful sign-in, which then redirects back to Design, which immediately re-initiates the login prompt. The loop never resolves, effectively locking the user out of the product entirely despite holding a valid session.
The issue is not isolated to Claude Design specifically. Research context reveals that login loop failures have been documented across multiple Anthropic products and surfaces, including the Claude.ai web platform, the Claude Code VS Code extension, and the Claude Desktop App on macOS. On Claude.ai, users have reported being unable to progress past the login screen when authenticating via Google or email, with the session failing to persist and cycling back to the start. Within Claude Code, similar OAuth-related infinite loops have appeared during the `/login` flow via Anthropic Console, and the Claude Desktop App has separately exhibited "Invalid authorization" errors on launch that prevent authentication from completing. The breadth of these reports across platforms suggests the problem is not isolated to a single interface but may reflect systemic issues in Anthropic's session management or OAuth token handling infrastructure.
From a technical standpoint, redirect loop failures of this kind typically stem from mismatched session tokens, misconfigured OAuth callback URLs, or cookie-scoping problems that cause the server to reject an ostensibly valid session and re-initiate the auth flow. When a web app and an identity provider disagree about the state of a session — for instance, because a cookie is scoped too narrowly or a token has been invalidated server-side — the result is precisely the bouncing behavior described. For a product like Claude Design, which appears to be a distinct route or application context within the claude.ai ecosystem, the failure may also indicate that Design-specific authentication middleware is not correctly inheriting or recognizing sessions established at the top-level claude.ai domain.
The broader significance of these authentication failures extends beyond user inconvenience. Anthropic has been aggressively expanding its product surface in 2025 and into 2026, with Claude powering an increasing number of discrete tools, integrations, and specialized interfaces. As the company pushes into design tooling, coding assistants, and desktop applications simultaneously, the authentication layer becomes a critical shared dependency. Failures in that layer compound across products, and the absence of official fixes or status-page acknowledgments noted in the research context suggests that Anthropic's incident response and public communication around infrastructure issues may not yet be scaling proportionally with its product growth. Until official remediation is provided, affected users are generally advised to clear browser cache and cookies, attempt access via incognito mode, or consult Anthropic's status page for service degradation notices.
Read original article →