Systematic Service Failures - $440 Refund Request After Month of Critical Bugs (WSL2)

Found 3 possible duplicate issues:

1. https://github.com/anthropics/claude-code/issues/9950
2. https://github.com/anthropics/claude-code/issues/11810
3. https://github.com/anthropics/claude-code/issues/14261

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

NOT a Duplicate - Comprehensive Refund Request Issue

This issue is NOT a duplicate of the suggested issues. Here's why:

Comparison

| Suggested Issue | Topic | This Issue (#21601) |
|----------------|-------|---------------------|
| #9950 | Individual connection issue | $440 refund request across 3 distinct bugs over entire month |
| #11810 | Specific error report | Systematic service failure with financial impact documentation |
| #14261 | Single bug report | Comprehensive case combining #18762, #20756, #21417 + refund demand |

Why This Issue is Unique

This is not a bug report. This is a formal refund request and service failure documentation that:

1. Combines 3 separate critical bugs (#18762, #20756, #21417) into one comprehensive case
2. Documents $440 in subscription fees for unusable service
3. Documents 55+ hours of time lost troubleshooting Anthropic's bugs
4. Records 5 failed support attempts over 10+ days with only AI bot responses
5. References Anthropic's own acknowledgment of "multiple service disruptions throughout January 2026" (email from support@mail.anthropic.com, Jan 29, 2026)

Anthropic Has Already Acknowledged the Problem

From Anthropic Support email (Jan 29, 2026):

"I can confirm that there were indeed multiple service disruptions throughout January 2026 that affected Claude Pro and our API services."

This Issue Needs Human Attention

No automated bot can process a refund request. This requires a human from Anthropic's billing or support team to:

1. Review the documented service failures
2. Process the $440 refund
3. Respond to the related GitHub issues (#18762, #20756, #21417)

Please keep this issue OPEN.

---
Reporter: Lee Jyong Chul (이종철)
Company: Whitehat Marketing Co., Ltd.
Email: jyongchul@naver.com

Update: Both Refund Requests REJECTED (January 29, 2026)

Following Anthropic's own email instructions to use the web refund process, I submitted refund requests for both accounts. Both were automatically rejected.

Results

| Account | Status | Rejection Reason |
|---------|--------|-----------------|
| jyongchul@naver.com ($220) | ❌ REJECTED | "Previously issued refund" |
| jyongchul@gmail.com ($200) | ❌ REJECTED | "Previously issued refund" |

The Contradiction

1. Anthropic's own support email (Jan 29) acknowledged "multiple service disruptions throughout January 2026"
2. Anthropic's own support email directed me to the web refund process
3. The web refund process rejected both requests with an automated policy

So Anthropic:

• ✅ Acknowledges the service was broken
• ✅ Directs me to request a refund
• ❌ Then automatically rejects the refund

This is a Catch-22 by design: acknowledge the problem, point to a process, but make the process automatically reject.

Previous Refunds Were for DIFFERENT Issues

Any previous refunds were for earlier, separate incidents — NOT for January 2026's month-long systematic failures across 3 critical bugs (#18762, #20756, #21417).

A prior refund for a prior problem does not negate a new refund for a new problem. Each billing period is a separate service contract.

Next Steps

Since Anthropic has:

• ❌ Provided zero human responses (only AI bots)
• ❌ Rejected legitimate refund requests via their own recommended process
• ❌ Failed to respond to any of the 3 GitHub issues (#18762, #20756, #21417)

I will now proceed with:

1. Credit card chargeback ("services not rendered") for both accounts
2. Public disclosure on Reddit, Hacker News, Twitter/X
3. Consumer protection complaints (Korea Consumer Agency, FTC)

---
Reporter: Lee Jyong Chul (이종철)
Company: Whitehat Marketing Co., Ltd.

Update: Chargeback Notice Sent (January 29, 2026)

Formal chargeback notice has been sent to support@anthropic.com and billing@anthropic.com.

Summary of All Failed Resolution Attempts

| # | Date | Action | Result |
|---|------|--------|--------|
| 1 | Jan 25 | GitHub #20756 (OAuth broken) | No human response |
| 2 | Jan 28 | GitHub #21417 (API errors) | Bot auto-reply, threatened auto-close |
| 3 | Jan 29 | Email refund request (both accounts) | AI bot response only |
| 4 | Jan 29 | Web refund process (both accounts) | Auto-rejected ("previously issued refund") |
| 5 | Jan 29 | GitHub #21601 (comprehensive case) | Bot auto-reply, threatened auto-close |
| 6 | Jan 29 | Chargeback notice email | Sent to support + billing |

The Contradiction

• Anthropic acknowledges "multiple service disruptions throughout January 2026"
• Anthropic directs me to web refund process
• Web process rejects refund citing prior refund (different period, different issues)
• Result: No path to resolution exists within Anthropic's system

48-Hour Window

Anthropic has until January 31, 2026 23:59 KST to issue direct refunds ($420 total) before chargebacks are filed.

A direct refund is better for both parties. Chargebacks carry processing fees and merchant rating impact.

Final Status Update — January 29, 2026

Email Response Analysis Complete

| Inbox | Total Anthropic Emails | Human Responses |
|-------|----------------------|-----------------|
| Naver (jyongchul@naver.com) | 22 emails since Jan 28 | 0 |
| Gmail (jyongchul@gmail.com) | 6 emails since Jan 28 | 0 |
| Total | 28 emails | 0 human responses |

Every single response was from "Fin AI Agent" — automated bot replies only.

Bot Response Pattern

1. Generic refund instructions → "Use web refund process" (which auto-rejects)
2. Acknowledgment template → "Thank you for reaching out, we'll be with you soon" (never happened)
3. Auto-close → "We haven't heard from you in a while" (closing tickets without resolution)
4. Chargeback warning → "If you file disputes, we cannot process direct refunds" (catch-22)

Card Information Confirmed

Both accounts use the same Mastercard ending in 2238:

• Account 1 (Naver): Direct Mastercard charge — $220
• Account 2 (Gmail): Stripe Link → same Mastercard 2238 — $220
• Total: $440

Next Steps

1. Credit card chargeback — filing with card company for both charges ($440 total) citing "Services Not Rendered"
2. Consumer protection filing — Korea Consumer Agency (한국소비자원)
3. Public disclosure — Reddit, Hacker News, Twitter (all content prepared)

After 28 automated emails and 0 human responses over 10+ days, there is no remaining private channel to resolve this.

---

⚠️ DO NOT auto-close this issue. This is an active consumer dispute with documented evidence of systematic refund avoidance.

Status Update — January 30, 2026 (Evening KST)

Actions Taken Today

1. Samsung Card (삼성카드) chargeback formally filed — Electronic complaint submitted via samsungcard.com requesting expedited processing of $440 dispute (Mastercard 2238)

1. Final protest email sent to Anthropic — Comprehensive email with real terminal logs showing API errors throughout January, sent to support@anthropic.com and billing@anthropic.com

1. Second account protest sent — Separate email confirming Gmail account (jyongchul@gmail.com, $220) demands are identical

1. GitHub issues #20756 and #21417 updated with chargeback filing notice

Anthropic Response Summary (as of Jan 30 evening)

| Channel | Attempts | Human Responses |
|---------|----------|-----------------|
| Email (support@anthropic.com) | 22+ | 0 |
| Web refund process | 2 (both accounts) | Auto-rejected |
| GitHub issues (4 issues) | 9 total | 0 |
| Total human responses | | 0 |

Latest Anthropic email (Jan 30, 10:34 UTC): Generic auto-reply template — "Thank you for reaching out... A member of our team will be with you as soon as we can."

Current Status

• ⏳ Samsung Card processing chargeback ($440)
• ⏳ Awaiting Samsung Card callback (1-3 business days)
• ❌ Still zero human responses from Anthropic after 22+ attempts
• ✅ All evidence documented and submitted

For Anthropic Staff

This can still be resolved directly. Issue a $440 refund ($220 × 2 accounts) and I will withdraw the chargeback and remove public complaints. Your own support acknowledged "multiple service disruptions throughout January 2026."

Contact: jyongchul@naver.com / jyongchul@gmail.com / +82-10-9333-2028

Update: Jan 31, 2026 — Email Server Blocking + Chargeback Filed

1. Anthropic Email Server Now Blocking Customer Emails

Emails to Anthropic support are being rejected with:

550 5.7.1 IP address blacklisted by recipient

Customer cannot reach support through email. Combined with zero human responses across all channels.

2. Samsung Card Chargeback Filed (Jan 30)

• Card: Mastercard ending 2238
• Amount: $440 ($220 x 2 accounts)
• Reason: Service unusable ~90% of January 2026, no support response

3. GitHub Issues Update

• #21692: opusplan model triggers extreme API connection errors (new finding)
• #20756: OAuth still broken in WSL2
• #21417: IP-change-correlated connection drops

All channels exhausted. Awaiting chargeback resolution.

Update: Jan 31, 2026 — Complete January 2026 Failure Evidence

Email Server Blocking

Anthropic's Intercom mail server now blacklists customer IPs:

550 5.7.1 IP address blacklisted by recipient

Gmail Daily Sending Limit

After sending 22+ emails requesting a human response, Gmail has also rate-limited my sending. This means Anthropic's non-responsiveness has exhausted my ability to even contact them.

Full January 2026 Error Log (Partial Sample)

From actual Claude Code sessions across multiple projects, computers, and accounts:

• API Error: Connection error → 100+ occurrences across 30+ work sessions
• Request timed out. Check your internet connection → Retries fail up to 10/10
• OAuth completely broken → /login fails with "OAuth error:" (Issue #20756)
• opusplan model → Near-constant connection errors (Issue #21692)
• MCP 16+ hour hang (Issue #15945)
• .claude.json corruption — auto-locked (Issue #18998)
• Config drift across sessions (Issue #18762)

Why Two Accounts ($220 × 2 = $440)

I opened a second account (jyongchul@gmail.com) believing the errors were account-specific. The same errors occurred on both accounts, proving this is a systemic Anthropic infrastructure problem, not a user issue.

Anthropic's Own Admission

"I can confirm that there were indeed multiple service disruptions throughout January 2026 that affected Claude Pro and our API services." — Fin AI Agent from Anthropic, January 29, 2026

Current Status

• Samsung Card chargeback: Filed Jan 30, 2026
• Waiting for Samsung Card to process
• Evidence package prepared and submitted
• 22+ email attempts, 9 GitHub issues, ZERO human responses from Anthropic

All Related Issues

• #21692 — opusplan API connection errors + Docker conflicts
• #20756 — OAuth broken in WSL2
• #21417 — API connection errors correlated with IP changes
• #20584 — API timeout despite healthy resources
• #18998 — .claude.json corruption
• #18762 — Plugin-MCP config drift
• #17935 — API timeout after successful tool execution
• #15945 — MCP 16+ hour hang

I wasted my entire January 2026 dealing with Anthropic's broken service. $440 for a service that was ~90% unusable.

CRITICAL EVIDENCE: 66 Official Anthropic Incident Emails in January 2026

I have compiled all 66 incident notification emails sent by Anthropic's own status system during January 2026, proving systematic service failures across 14 separate days:

Incident Timeline (From Anthropic's Own Emails)

| Date | # Emails | Incident Description |
|------|----------|---------------------|
| Jan 6 | 1 | Errors on requests to Haiku 3.5 |
| Jan 7 | 4 | Elevated error rates on Opus 4.5 |
| Jan 9 | 3 | Increased latency on Sonnet 4.5 |
| Jan 10 | 6 | Degraded performance for API requests (structured output) + Elevated errors on Sonnet 4.5 |
| Jan 11 | 3 | Elevated error rates on Sonnet 4.5 1M context |
| Jan 13 | 11 | Elevated errors for multiple models + Increased errors Opus 4.5 |
| Jan 14 | 13 | Elevated error rates on Opus 4.5 AND Sonnet 4.5 |
| Jan 15 | 1 | Elevated error rates on Opus 4.5 and Sonnet 4.5 (continued) |
| Jan 21 | 5 | Elevated errors on Opus 4.5, Sonnet 4.5, and Sonnet 3.5 |
| Jan 22 | 4 | Elevated errors on Claude Sonnet 4.5 |
| Jan 23 | 11 | Elevated error rates on claude.ai AND API + Invalid signature in thinking block |
| Jan 25 | 2 | Increased rate of errors for Opus 4.5 |
| Jan 26 | 1 | Increased rate of errors for Opus 4.5 (continued) |
| Jan 27 | 1 | Elevated errors on Claude Haiku 3.5 |

Total: 66 incident emails across 14 days in a 27-day period (52% of days had incidents)

Models Affected

• Opus 4.5 (most frequent)
• Sonnet 4.5
• Sonnet 4.5 1M context
• Haiku 3.5
• Multiple models simultaneously
• claude.ai AND API simultaneously

Refund Request History (From Email Records)

1. Jan 19, 2026: Sent detailed refund request with full technical documentation
2. Jan 19, 2026: Received AI bot response directing to web refund process
3. Jan 19, 2026: Bot auto-closed conversation ("We haven't heard from you in a while") — 4 hours after sending
4. Jan 19-31: Web refund auto-rejected ("Previously issued refund")
5. Jan 19-31: 22+ follow-up emails sent, ZERO human responses

This Is Not a User Issue

• 66 incident emails from Anthropic's own system prove systematic infrastructure failures
• Both accounts (jyongchul@naver.com, jyongchul@gmail.com) affected identically
• $440 paid for service that was demonstrably broken ~50% of days
• Samsung Card chargeback filed Jan 30, 2026

All 66 incident emails are preserved as evidence for the Samsung Card chargeback dispute.

UPDATE (2026-02-01): Anthropic's Own Admission of Service Failure — Evidence for Samsung Card Chargeback

Samsung Card Chargeback Filed (Case TCKT1000160744)

I have filed a formal chargeback with Samsung Card (Mastercard) for $400 ($200 × 2 accounts). Samsung Card case number: TCKT1000160744 (filed 2026-02-01).

---

Anthropic's Official Admissions (Direct Quotes)

Anthropic's own support system sent the following admissions via email to jyongchul@naver.com:

Admission #1 (2026-01-29, from Fin AI Agent / support@mail.anthropic.com):

"I can confirm that there were indeed multiple service disruptions throughout January 2026 that affected Claude Pro and our API services."

Admission #2 (2026-01-29, from Anthropic Support):

"I can see that January 2026 was indeed a challenging month for our service. Multiple significant incidents occurred throughout the month... including elevated error rates on Claude models, authentication issues, and degraded performance."

These admissions directly contradict the earlier claim that the only issue was a "27-minute authentication incident on January 22."

---

Evidence Summary

| Evidence | Detail |
|----------|--------|
| Incident alert emails | 66 emails from incident-notifications@anthropic.com in January 2026 |
| Days with incidents | 14 out of 27 days (52%) |
| Anthropic admission | Service disruptions confirmed by Anthropic's own support |
| Refund requests sent | 22+ emails |
| Human responses | 0 (only AI bot auto-replies) |
| Web refund attempts | Both accounts auto-rejected ("Previously issued refund") |
| Email IP blocked | Since 2026-01-31 (SMTP 550 5.7.1 rejection) |
| GitHub issues filed | 9 issues, 0 human responses from Anthropic |

---

Demand

Since Anthropic has admitted the service failures but refuses to process refunds (0 human responses after 22+ requests, email IP blocked), I am pursuing this through:

1. ✅ Samsung Card Mastercard chargeback (filed 2026-02-01)
2. ✅ This public GitHub evidence record
3. Pending: Korean Consumer Agency complaint

$200 × 2 accounts = $400 total refund demanded.

The original .eml email files containing Anthropic's admissions are preserved and available for Samsung Card's review upon request.

New Evidence Update (February 1, 2026): Third-Party Cross-Platform Bug Confirmation

Independent Confirmation by @ghcreative869

A third-party user (@ghcreative869) has independently confirmed the same OAuth → Subscription linking bug on native Linux (Ubuntu 25.10), proving this is NOT a platform-specific issue but a systemic Anthropic backend defect.

Key quote from third-party report:

"The root cause appears to be in Anthropic's OAuth → Subscription verification pipeline, not platform-specific (WSL2 vs native Linux)"

"OAuth token generation succeeds ✓ / Subscription status lookup fails or returns incorrect data ✗ / hasExtraUsageEnabled: false is set by default ✗ / Requests are silently rejected → '0 tokens' or connection errors"

Source: https://github.com/anthropics/claude-code/issues/20756#issuecomment-3830143498

Why This Is Decisive Evidence

| # | Point | Significance |
|---|-------|-------------|
| 1 | Independent reproduction | Third party with no relation to me reproduced the same backend bug |
| 2 | Platform-independent | Occurs on both WSL2 AND native Linux → Anthropic server-side defect confirmed |
| 3 | Paid subscriber impact | hasExtraUsageEnabled: false incorrectly applied to valid Max subscribers ($220/mo) |
| 4 | Multiple reports | Same symptoms in #17966, #6008, #12020 |

Actions Taken Today (Feb 1, 2026)

1. ✅ Updated Samsung Card (Mastercard) chargeback evidence package with third-party confirmation
2. ✅ Submitted supplementary evidence to Samsung Card via online complaint (전자민원접수)
3. ✅ This GitHub update

Running Totals

• Refund requests sent: 22+
• Human responses from Anthropic: 0
• GitHub issues filed: 9
• Third-party confirmations: 2 (now cross-platform)
• Chargeback status: Filed with Samsung Card (Mastercard), supplementary evidence submitted
• Daily protest emails: Active (cron, 00:00 UTC)

@anthropics/claude-code-team: You now have independent third-party confirmation that your OAuth→Subscription pipeline is broken for paying subscribers across multiple platforms. This is a billing-impacting backend defect. $440 refund for two Max accounts (jyongchul@naver.com + jyongchul@gmail.com) is the minimum appropriate response.

New Evidence: Anthropic's Own AI Bot Admits Refund Eligibility (Feb 1, 2026)

Just hours after I submitted the third-party cross-platform evidence, Anthropic's Fin AI Agent responded with a significant admission:

"The technical issues you've documented may be relevant to your refund eligibility once the chargeback process allows for direct refund processing."

Source: Email from Fin AI Agent (support@mail.anthropic.com) → jyongchul@gmail.com, Feb 1, 2026 12:40 KST

Why This Matters

1. Anthropic's own system acknowledges refund eligibility based on the documented technical issues
2. Yet the bot demands I withdraw the chargeback first before processing — with no written guarantee the refund will be approved
3. This is a catch-22: 22+ refund attempts were ignored/rejected BEFORE I filed the chargeback. Withdrawing now with no guarantee would leave me with zero leverage.
4. Still Day 4 with ZERO human responses — only AI bot auto-replies

The Pattern

| Date | Channel | Response | Result |
|------|---------|----------|--------|
| Jan 19-31 | Email (22+) | AI bot / none | Refund refused |
| Jan 19 | Web refund | Auto-system | "Previously issued refund" — rejected |
| Jan 31 | Email | Server | IP blacklisted (550 5.7.1) |
| Feb 1 | Email | AI bot | Admits refund eligibility, demands chargeback withdrawal first |

Anthropic admits the technical issues justify a refund but refuses to connect me to a human who can actually process it. This is the definition of bad faith customer service.

Running total: Day 4, 24+ contact attempts, 0 human responses.

Samsung Card Chargeback Progress Update (Feb 2, 2026)

Samsung Card has assigned a dedicated human agent (고다영, d1226.ko@partner.samsung.com) to process the chargeback through Mastercard's dispute resolution.

Samsung Card's Confirmation:

• Both transactions identified: $200 (Jan 7, auth 393035) + $200 (Jan 11, auth 394016) = $400
• Dispute will be filed through Mastercard brand dispute resolution (취소매출 미접수 / Services Not Rendered)
• Estimated timeline: 45-75 business days per Mastercard procedures
• Evidence package submitted: 78 email files (30 refund requests sent, 43 AI bot responses received, 3 invoices, 1 summary document)

Evidence Submitted to Samsung Card:

1. All 30+ refund request emails sent to Anthropic (original .eml files)
2. Anthropic's admission: "multiple service disruptions throughout January 2026"
3. Anthropic's refund eligibility admission: "may be relevant to your refund eligibility"
4. All 43 AI bot responses (proving zero human responses)
5. Payment receipts/invoices
6. GitHub issues cross-references (#20756, #21417, #20584, #18762, #17935, #15945)
7. Third-party independent confirmation (@ghcreative869 on native Linux)

Status: Formal Mastercard dispute process initiated. Awaiting Anthropic's response through card network.

Update: API failures continue into February 2026

As of February 2, 2026 ~19:00 KST, API connection errors are occurring again. Claude Code is returning API Error: Connection error. on every command, with wait times of 3-14 minutes before each failure.

January failures have now extended into February. The $440 chargeback is filed with Samsung Card. Still zero human responses from Anthropic support.

Additional API Error Report (2026-02-03)

Another wave of API 500 errors just occurred:

API Error: 500 {"type":"error","error":{"type":"api_error","message":"Internal server error"},"request_id":"req_011CXmPdEfd5dHw7mvW2JGDw"}
API Error: 500 {"type":"error","error":{"type":"api_error","message":"Internal server error"},"request_id":"req_011CXmQ5okRgvzvePmWejTPE"}
API Error: 500 {"type":"error","error":{"type":"api_error","message":"Internal server error"},"request_id":"req_011CXmQ6djT27JGYvL5RWawb"}
API Error: 500 {"type":"error","error":{"type":"api_error","message":"Internal server error"},"request_id":"req_011CXmQ6ns7mYMAQAXG2ktD4"}
API Error: 500 {"type":"error","error":{"type":"api_error","message":"Internal server error"},"request_id":"req_011CXmQ77Uoi5Bb7fvtC65wB"}
API Error: 500 {"type":"error","error":{"type":"api_error","message":"Internal server error"},"request_id":"req_011CXmQ7GkAi8ZTFypSWggvK"}

Time: 2026-02-03 ~18:50 KST
Environment: WSL2, Claude Code
Pattern: 6 consecutive 500 errors, unable to continue work

This is ongoing evidence that Anthropic's API infrastructure remains unstable.

The chargeback dispute with Samsung Card is in progress. This additional failure will be reported to the card company as supplementary evidence of continued service failures.

---
Request IDs for Anthropic internal investigation:

• req_011CXmPdEfd5dHw7mvW2JGDw
• req_011CXmQ5okRgvzvePmWejTPE
• req_011CXmQ6djT27JGYvL5RWawb
• req_011CXmQ6ns7mYMAQAXG2ktD4
• req_011CXmQ77Uoi5Bb7fvtC65wB
• req_011CXmQ7GkAi8ZTFypSWggvK

Bro, you are angry lol, just chill.

they don't care about their customer, that's why they are treating you, me and others like that.

Stop losing energy, they'll have a backfire.

It's a shame, plain and simple. They bill you for a service they don't provide. That's it for me, bye bye anthropic.

Closing for now — inactive for too long. Please open a new issue if this is still relevant.

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.