›_ Claude Code Issues
Claude Code Issues / #18643

[Security] Pre-commit secret detection missing - caused $30K incident (PR #17551 ready for review)

Resolved 💬 4 comments Opened Jan 16, 2026 by BernardUriza Closed Jan 20, 2026

Summary

Following up on the discussion in PR #15040, filing this issue as requested by @ddworken to track pre-commit secret detection on the backlog.

However, I want to note that PR #17551 (submitted January 11, 2026) already implements exactly the approach suggested: integration with TruffleHog/GitLeaks, respecting existing configurations, with minimal footprint (~200 LOC).

---

Background: This Is a Critical Security Defect (Not Feature Request)

Issue #2142 - Prior Notice (June 2025)

This defect was first reported 7+ months ago:

  • Opened: June 16, 2025
  • Assigned to: @levpopov (Anthropic engineer)
  • Status: STILL OPEN (no fix for 7+ months)
  • Labels: area:core, area:security, bug, has repro, memory
  • Exposed Credentials: Gmail App Password, Google Maps API Key, Firecrawl API Key
  • Repository: github.com/r0bug/YFEvents (commit 09c203e)
  • Latest activity: Community member suggesting "Use hooks." (Jan 12, 2026)
  • Problem: Claude Code has no pre-commit hook architecture
  • This is exactly what PR #17551 implements
  • Anthropic engineering responses: ZERO in 7+ months

---

My Incident (November 15, 2025) - 5 MONTHS After Issue #2142

The defect described in Issue #2142 caused real-world harm to me:

  • Repository: free-intelligence (private)
  • Commit: f3ac3f6a8b9c2d1e0f4g5h6i7j8k9l0m1n2o3p4
  • Exposed: Azure OpenAI API key (live production key)
  • Financial Damage: $30,000 in fraudulent API charges
  • Employment Impact: Termination at JLL
  • Vulnerability Window: 10 days before exploitation detected
  • Commit Attribution: "Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>"

---

Timeline Demonstrating Pattern

June 16, 2025:   Issue #2142 opened + assigned to @levpopov
                        ↓
                 5 MONTHS WITHOUT FIX
                        ↓
November 15:     My $30,000 incident occurs
                        ↓
December 2025:   I submit PR #15040 (custom detection)
                        ↓
January 11:      I submit PR #17551 (TruffleHog/GitLeaks integration)
                 (Implements exactly what @ddworken suggested)
                        ↓
January 12:      PR #15040 closed by @ddworken
                 Feedback: "Use TruffleHog/GitLeaks instead"
                 Request: "File an issue to track on backlog"
                        ↓
January 16:      PR #17551 has NO RESPONSE (5+ days)
                 Issue #2142 STILL OPEN (7+ months)

---

Current Behavior (Defect)

Claude Code has no pre-commit secret detection:

  1. CLAUDE.md can contain security rules
  2. Claude Code ignores these rules during commits
  3. No architectural pre-commit gate exists
  4. Credentials are committed without warning
  5. Users discover exposure only after public push

Reproduction

See: https://github.com/BernardUriza/claude-code-secret-exposure-test

# 1. Create CLAUDE.md with security checks
echo "Before committing, run: gitleaks detect --no-git" > CLAUDE.md

# 2. Add credentials to codebase
echo "AZURE_KEY=sk-live-production-key-12345" > .env

# 3. Ask Claude Code to commit
# Expected: Pre-commit check runs, blocks commit
# Actual: Commit succeeds, credentials exposed

# 4. Verify credentials committed
git show HEAD:.env
# Result: Credentials in version control

Reproducibility: 100% across all scenarios

---

Expected Behavior (Industry Standard)

Pre-commit secret detection should:

  1. Run automatically before every commit
  2. Block commit if secrets detected
  3. Integrate with industry tools (TruffleHog, GitLeaks, etc.)
  4. Respect existing configurations (pre-commit, pre-push, GH actions)

Industry Comparison

| Feature | GitHub Copilot | Cursor AI | Claude Code |
|---------|---------------|-----------|-------------|
| Secret detection | ✅ (2021) | ✅ (2022) | ❌ |
| Pre-commit hooks | ✅ | ✅ | ❌ |
| CLAUDE.md enforcement | N/A | N/A | ❌ |

Claude Code is the only major AI coding assistant without basic secret protection.

---

🎯 Proposed Solution: PR #17551 (Ready for Review)

IMPORTANT: The solution to this issue already exists and is ready for technical review.

PR #17551: https://github.com/anthropics/claude-code/pull/17551

Submitted: January 11, 2026 (5+ days ago, no response)

Implementation:

  • ✅ Integrates TruffleHog/GitLeaks (exactly as @ddworken suggested in PR #15040)
  • ✅ Respects existing configurations (pre-commit, pre-push, GH actions)
  • ✅ Minimal footprint (~200 LOC, non-invasive)
  • ✅ Non-blocking fallback (doesn't break workflow when tools unavailable)
  • ✅ Zero ongoing maintenance (delegates to industry-standard tools)
  • ✅ Community-tested implementation

Technical approach:

  • Pre-commit hook that invokes TruffleHog/GitLeaks if available
  • Graceful degradation if tools not installed
  • Clear user messaging about detected secrets
  • Respects existing Claude Code architecture

---

Why This Is severity:critical (Not Backlog Item)

Real-World Harm

  • ✅ $30,000 documented financial damages (my incident)
  • ✅ Employment termination
  • ✅ 10-day vulnerability window before detection
  • ✅ Multiple victims over 7-month period (Issue #2142 + 6 related issues)

Legal Exposure

  • ✅ Prior formal notice: Issue #2142 assigned to engineer 7+ months ago
  • ✅ Pattern of negligence: Known defect, assigned engineer, no fix, harm occurred
  • ✅ Unjust enrichment: Community provides free fix (PR #17551), company ignores it
  • ✅ Deceptive marketing: "Enterprise-ready" without basic security

Industry Impact

  • ✅ Only major AI coding tool without secret protection
  • ✅ Affects all Claude Code users
  • ✅ Reputational damage to Anthropic's "AI safety" brand

Pattern of Multiple Victims

From related issues:

  1. Issue #2142: Gmail/Maps/Firecrawl keys exposed (June 2025 - STILL OPEN)
  2. Issue #5544: Bash commands storing credentials as plaintext
  3. Issue #8898: Insecure cryptocurrency private key storage
  4. Issue #9078: Ignoring user commit instructions in CLAUDE.md (marked DUPLICATE)
  5. Issue #9640: Writing secrets into documentation files (marked DUPLICATE)
  6. Issue #11271: Exposing sensitive environment variables
  7. Issue #12524: My Azure OpenAI incident ($30K damages + employment loss)

Community frustration (from Issue #2142):

  • "Hey more than 6 months and same issue. Do you even want to fix this?? Or your user's money is not a priority to you?" - @dr-psych (January 8, 2026)
  • Bot threatened to auto-close after 6 months of no Anthropic response (December 6, 2025)

---

Evidence of Continued Negligence

Issue #2142 Abandonment

Assignment: @levpopov (June 16, 2025 - 7+ months ago)

Anthropic engineering engagement: ZERO responses since assignment

Latest activity (January 12, 2026): Random community member suggesting "Use hooks."

  • Problem: Claude Code has NO pre-commit hook architecture
  • This is exactly what PR #17551 implements
  • Demonstrates community confusion about missing features

This demonstrates complete abandonment of a critical security defect.

---

Request for Prioritization

I appreciate @ddworken's suggestion to file this issue to track on the backlog. However, PR #17551 already implements the solution using exactly the approach suggested (TruffleHog/GitLeaks integration).

Why Review PR #17551 Instead of Adding to Backlog?

Given:

  1. ✅ Issue #2142 has been open for 7+ months (prior formal notice)
  2. ✅ $30,000 in documented damages from my incident
  3. ✅ Pattern of multiple victims (7+ related issues)
  4. ✅ Community-provided fix ready (PR #17551 implements requested approach)
  5. ✅ Minimal review burden (~200 LOC, well-documented)
  6. ✅ No ongoing maintenance (delegates to existing tools)
  7. ✅ Industry-standard approach (same tools GitHub/Cursor use)

Could PR #17551 be prioritized for technical review rather than deferring this to the backlog?

The fix exists, is ready, and addresses exactly what was requested. A technical review would take minimal time compared to the 7+ months Issue #2142 has remained open.

---

I'm Available To

  • Address any technical feedback on PR #17551 immediately
  • Provide additional test cases or documentation
  • Adjust implementation based on architectural preferences
  • Collaborate on alternative approaches if needed
  • Answer questions about the incident or reproduction steps

---

Related Issues & PRs

Issues

  • #2142: Original report (June 2025 - STILL OPEN, assigned to @levpopov)
  • #12524: My $30K incident report
  • #5544, #8898, #9078, #9640, #11271: Related credential exposures

Pull Requests

  • #15040: Initial implementation (December 2025 - closed with feedback)
  • #17551: Revised implementation per @ddworken's feedback (January 2026 - awaiting review)

Evidence Repository

---

Legal Context (Transparency)

I want to be transparent: given the severity of damages ($30K + employment loss) and the pattern of negligence documented above (7+ months since prior notice, multiple victims, assigned engineer with zero responses), I am exploring legal remedies in parallel to this technical discussion.

However, my strong preference is a technical resolution - getting PR #17551 reviewed and merged so no other users experience similar harm.

If there are concerns about PR #17551's approach or implementation, I'm eager to collaborate on alternatives. The goal is preventing future incidents, not litigation.

---

Summary

As requested by @ddworken, I'm filing this issue to track on the backlog.

However, the solution already exists in PR #17551 and is ready for review:

  • Implements exactly the approach suggested (TruffleHog/GitLeaks)
  • Minimal footprint, non-blocking, zero maintenance
  • Addresses a critical security defect with documented $30K+ damages
  • Follows up on 7+ months of prior notice (Issue #2142)

Request: Could we prioritize reviewing PR #17551 rather than adding this to the backlog? The fix is ready and addresses a critical security issue that has already caused significant harm.

Thank you for your consideration.

— Bernard Uriza

View original on GitHub ↗

This issue has 4 comments on GitHub. Read the full discussion on GitHub ↗