[BUG] Cowork broken since April 15: 503 on /v1/environment_providers/private/.../environments endpoint, persistent 12+ days, full reauth doesn't resolve

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?

Summary

Cowork has been completely non-functional on my Pro account since April 15, 2026, 04:20:54 AM CT. Every Cowork task fails. Regular Claude chat works fine. Full reauth, reinstall, network changes do not resolve.

The smoking gun in the logs is repeated 503 errors on a specific backend endpoint:
GET /v1/environment_providers/private/organizations/{org_id}/environments
→ 503 "upstream connect error or disconnect/reset before headers.
reset reason: remote connection failure,
transport failure reason: delayed connect error: Connection refused"

This is the Cowork sandbox provisioning endpoint. When it returns 503, no execution environment can be created, which is why tasks accept the connection then die mid-flight.

Note on repo and related issues

This bug is in the Cowork sandbox provisioner, not Claude Code CLI. Filing here based on precedent: #48310 (Cowork tabs missing post-April 15 update), #52367 (Dispatch broken since April 15 outage, Max user, no recovery), #53442 (Cowork Drive MCP) — all area:cowork bugs accepted in this repo. There is no separate Cowork repo.

This appears to be part of a wider post-April 15, 2026 service degradation pattern. The April 15 timing in this report matches the outage referenced in #52367 ("Anthropic outage on Wednesday, April 15, 2026"). My account is one of the ones that never recovered.

Environment

• Plan: Claude Pro
• OS: macOS (Mac)
• Client: Claude Desktop
• Org ID: cdb35e23-a32f-47de-8084-eb4b39448be7
• Routing path: /environment_providers/private/... (private path, not public)

Timeline (from ~/Library/Logs/Claude/claude.ai-web.log)

| Date | Errors logged | Notes |
|---|---|---|
| Apr 12–14 | 1–5/day | Background noise only |
| Apr 15 | 72 | First 503 spike begins 04:20:54 AM on environments endpoint |
| Apr 16 | 3 | Quiet |
| Apr 18 | 223 | 190 "Failed to fetch" — stopped using Cowork after this day |
| Apr 19 | 1 | Not used |
| Apr 20 | 13 | First [COMPLETION] Request failed errors at 14:56 |
| Apr 21–26 | minimal | Stopped using |
| Apr 27 | 127 | Confirmed still broken on retry today |

Troubleshooting already attempted

• Full reauth (sign out everywhere, kill processes, delete ~/Library/Application Support/Claude and ~/Library/Logs/Claude, reinstall, sign back in) — multiple times
• VPN fully quit
• App fully quit and relaunched
• System reboots
• Email support — multiple attempts, no response

Why this is not the same as #45175

#45175 describes mid-task severance from VM startup. This issue is different:

• Failure originates upstream of the client (503 from Anthropic's environment provisioner)
• Specific to the private routing path on this org ID
• Persistent for 12+ days, not intermittent
• No client-side fix resolves it

Diagnosis

The endpoint /v1/environment_providers/private/organizations/{org_id}/environments is returning 503 with "Connection refused" on the upstream. This suggests either:

1. A backend service for the private environment provisioner failing to connect to its upstream
2. Org-specific routing/config issue on this org_id
3. A deploy on/around April 15 that broke the private path for some accounts

Local fixes cannot resolve this. Needs backend investigation.

What I need

• Confirmation this is being looked at on the backend
• Workaround if any exists
• Refund credit for the period Cowork has been unusable (paying for Pro Cowork access since April 15)

What Should Happen?

Cowork tasks should successfully provision an execution environment and complete normally.

Instead: every Cowork task fails. The connection is accepted, the task starts ("Working on it..."), then the request dies mid-flight. The root cause appears to be 503 "upstream Connection refused" errors on the Cowork environment provisioning endpoint, which has been failing for my organization since April 15, 2026.

Regular Claude chat works fine. Only Cowork is affected.

Error Messages/Logs

2026-04-15 04:20:54 [error] [REACT_QUERY_CLIENT] QueryClient error:
{"type":"api_error","statusCode":503,
"endpoint":"/v1/environment_providers/private/organizations/cdb35e23-a32f-47de-8084-eb4b39448be7/environments",
"method":"GET",
"message":"upstream connect error or disconnect/reset before headers.
reset reason: remote connection failure,
transport failure reason: delayed connect error: Connection refused"}

2026-04-20 14:56:28 [error] [COMPLETION] Request failed
{"name":"TypeError","message":"Failed to fetch"}
2026-04-20 14:56:28 [error] [COMPLETION] Not retryable error, throwing
2026-04-20 14:56:28 [error] [COMPLETION] Non-API stream error
{"name":"TypeError","message":"Failed to fetch"}

2026-04-20 14:54:38 [warn] MaxListenersExceededWarning:
Possible EventEmitter memory leak detected. 11 ...AutoUpdater_$_updaterState
listeners added.

Error count timeline (from ~/Library/Logs/Claude/claude.ai-web.log):
- April 12-14: 1-5 errors/day (background)
- April 15: 72 errors (FIRST 503 SPIKE at 04:20:54 AM)
- April 18: 223 errors (190 "Failed to fetch")
- April 20: 13 errors (first [COMPLETION] failures)
- April 27: 127 errors (still broken on retry today)

Steps to Reproduce

1. Sign into Claude Desktop on macOS as a Pro user with org_id cdb35e23-a32f-47de-8084-eb4b39448be7
2. Attempt any Cowork task (file creation, code execution, multi-step workflow)
3. Observe: connection accepts, task begins ("Working on it..."), then severs mid-flight
4. Check ~/Library/Logs/Claude/claude.ai-web.log for 503 errors on /v1/environment_providers/private/organizations/{org_id}/environments

Note: this reproduces 100% of the time on my account since April 15. Other Pro accounts may not reproduce, suggesting an org-specific routing or config issue on Anthropic's backend.

Mitigations attempted (none resolved the issue):

• Full reauth: sign out everywhere, kill processes, delete ~/Library/Application Support/Claude and ~/Library/Logs/Claude, reinstall, sign back in
• VPN fully quit
• App fully quit and relaunched
• System reboots
• Email support (multiple attempts, no response)

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

Whatever Claude Desktop version was current on April 14, 2026. Cowork worked normally up until April 15, 04:20 AM CT, when 503 errors began on the environment provisioning endpoint.

Claude Code Version

Claude 1.3883.0

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

This bug is fundamentally about the Cowork sandbox provisioner, not the Claude Code CLI. The Terminal/Shell and Platform fields above don't really apply but are required fields, so I've selected the closest match. The actual reproduction is in Claude Desktop's Cowork mode on macOS.

Logs available on request — full ~/Library/Logs/Claude/claude.ai-web.log spans April 11-27 with detailed error timestamps.