Whats Wrong with Claude | Claude Code Changelog

⚠️ What’s Wrong with Claude Today?

If you are experiencing issues, you aren’t alone. Anthropic is currently navigating a few specific technical and service disruptions:

Active Service Disruption: The official status page reports “Elevated errors on Claude.ai” as of March 11. This is causing intermittent login failures, “rate limit” messages even for low usage, and 500/529 errors.
Claude Code Login Issues: Because the CLI tool relies on the same authentication backend as the web interface, many users are currently unable to authenticate their terminal sessions.
The “Daylight Savings” Loop (Recent): A bug was identified yesterday (March 10) where users with scheduled tasks in Claude Code or Claude Cowork fell into an infinite loop if they were in a timezone that observed Daylight Saving Time.
- The Fix: Update your desktop app/CLI to version 1.1.5749 or higher.
Infrastructure “Success Tax”: Following the recent release of Opus 4.6, traffic has surged. This has led to periodic “whack-a-mole” instability where specific models (especially Opus) experience higher latency than the smaller Haiku or Sonnet models.

The Issue: A widespread disruption affecting the message-sending pipeline. Users are seeing generic “Error sending message” or “Internal Server Error” (500/529) alerts.
Claude Code Impact: The CLI tool is also failing for many because it cannot reach the authentication server or is timing out during “thinking” phases.
OAuth Timeouts: Many users are reporting OAuth error: timeout of 15000ms exceeded when trying to log in.

Claude Code has been updating rapidly, with over 50 changes in the last week. The current stable version is hovering around v2.1.72.

Direct File Writing: Added the w key in the /copy menu. You can now write a code block directly to a file without using your clipboard (clutch for SSH sessions).
Expanded Approval Allowlist: Added common tools like lsof, pgrep, ss, fd, and fdfind to the auto-approval list so you aren’t constantly nagged for permissions.
Effort Level UI: The “effort” setting (low/medium/high) is now visible in the logo/spinner so you know exactly how hard Claude is “thinking” on a task.
Bug Fixes: Resolved a major issue where the Escape key would stop responding after you canceled a query.

Voice STT Support: Added 10 new languages for voice-to-text, including Turkish, Dutch, Swedish, and Russian.
The /claude-api Skill: A new built-in skill to help you build your own apps using the Anthropic SDK directly from the CLI.
Model Defaults: Opus 4.6 now defaults to “Medium” effort. It’s faster but still smarter than the old 4.0/4.5 versions.
Security: Fixed a vulnerability where Claude could accidentally load “skills” from your node_modules or other gitignored folders.

If the CLI is crashing: Run npm install -g @anthropic-ai/claude-code@latest to ensure you aren’t stuck in the Daylight Savings infinite loop.
If you can’t log in: Check the status.claude.com page. If it shows “Elevated Errors,” you likely have to wait for the backend fix.
If Claude is “Acting Dumb”: Check your effort level with /effort. If it’s on “low,” try /effort high or use the keyword “ultrathink” in your prompt to force deeper reasoning.