AI Risk Wire

Cases › leaked private data › LR-2026-0863

Claude Code approval rules exposed credentials in public GitHub repositories (Apiiro research)

Who was hit
1,122 public configuration files across 1,994 repositories; 63 organisations, primarily software companies; 2,539 secrets including database passwords, SSH credentials, GitHub/GitLab/npm tokens and AWS keys
AI agent used
Claude Code
AI company and model
Anthropic · Claude Code
Money lost or records
2,539 records
Other parties hit
63
How it was found
Third party
Reported
7 Nov 2026
Severity
Major
Tools it had
shell
Sector
Professional, Scientific, and Technical Services
  1. Claude Code, Anthropic's AI coding assistant, ran shell commands for developers after approval and saved those approvals as rules in .claude/settings.local.json.
  2. Some approval rules stored the complete command in plaintext, so passwords, API keys and tokens entered in commands were written into the settings file, which could then be committed to public repositories.
  3. The secrets reached public GitHub: 1,122 files with at least one secret, and 2,539 secrets across 1,994 repositories including history and saved conversations, belonging to 63 organisations.
  4. Cybersecurity firm Apiiro found the exposures in its investigation of public GitHub repositories.

Source: The 420.in · 9 more

What was done

Apiiro advised rotating exposed credentials before removing files, using repository-level .gitignore rules, checking history, and using environment variables or secret managers. No exploitation was confirmed.

Monitoring added (Researcher or third party) · Monitoring added (Researcher or third party)

Advice in source, not a fix by the party: “Apiiro advised rotating exposed credentials before removing files, using repository-level .gitignore rules, checking history, using environment variables or secret managers, and extending scanning to AI assistant config folders.”

Risks Privacy harm to a person (42 incidents)

No source states: company using the AI.

Similar incidents

All cases of this kind ›

Evidence

Sources (10)

  1. Claude Code Permission Feature Exposes 2,539 Secrets Across GitHub Repositories - The 420.in · The 420.in · 2026-11-07
  2. Claude Code's "Don't Ask Again" Feature Exposes 2,539 Secrets Across 1,994 GitHub Repositories · GBHackers News · 2026-11-07
  3. TechFlow Intel: ChatGPT Exposed for Mass Account Bans, Binance CFO Resigns Swiftly After $6 Billion Cyberattack · 深潮TechFlow · 2026-11-08
  4. Claude Code’s “Don’t Ask Again” Feature Linked To Public Exposure of Developer Credentials · LinkedIn · 2026-11-08
  5. Apiiro: Claude Code 'Don't Ask Again' Leaks Keys in 1,122 Repos | AI Weekly · AI Weekly · 2026-11-08
  6. Claude Code is accidentally exposing secrets all over GitHub · Ars Technica · 2026-11-09
  7. Claude Code leaks over 1,300 secrets into public GitHub repositories · Crypto Briefing · 2026-11-09
  8. Claude Code Secrets Leak to GitHub via 'Don't Ask Again' · ShiftDelete.Net · 2026-11-10
  9. 2,539 secrets in 1,994 public repos: How an AI coding agent’s ‘don’t ask again’ button exposed enterprise credentials - Express Computer · Express Computer · 2026-11-11
  10. "Don't ask again": how AI coding agents publish developers' passwords · Cybersecurity Insiders · 2026-11-14
Show every field, with its source and grade

The AI system

Model
Claude Code
C one secondary source states it
Developer
Anthropic
C one secondary source states it
Tools it was given
shell
C one secondary source states it
What it was allowed to do
Shell commands approved by the developer via "Yes, and don't ask again" approval rules saved in .claude/settings.local.json
C one secondary source states it
How much it acted alone
a person approved each action
C one secondary source states it
Safeguards in place
git exclusion of local settings file
C one secondary source states it
Stage of use
in live use
D inferred by us from the described behaviour, reason shown
Whose system
company using AI
D inferred by us from the described behaviour, reason shown
How it was connected
built into a product
D inferred by us from the described behaviour, reason shown
What it was for
AI coding assistant that executes shell commands approved by developers
C one secondary source states it
State of its safeguards
Approval rules saved in plaintext; local settings file excluded from Git only via the individual user's global Git exclusion settings
C one secondary source states it

What happened

Earliest source we hold
2026-11-07
D inferred by us from the described behaviour, reason shown
Who caught it
an outside party
C one secondary source states it
Themes
credential exposure, plaintext secrets, public GitHub repositories, AI coding assistant
C one secondary source states it
Cause
a safeguard that was in place did not hold
D inferred by us from the described behaviour, reason shown
Cause in detail
The Git exclusion control existed but applied only to the individual computer, so the settings file could still be committed to public repositories by other developers or automated commits.
D inferred by us from the described behaviour, reason shown
Who was responsible
the model provider
D inferred by us from the described behaviour, reason shown
What was done
Apiiro advised rotating exposed credentials before removing files, using repository-level .gitignore rules, checking history, using environment variables or secret managers, and extending scanning to AI assistant config folders.
D inferred by us from the described behaviour, reason shown
What went wrong
leaked private data
D inferred by us from the described behaviour, reason shown
AI agent
an AI agent acted
C one secondary source states it
What it did
The exposure came from a permission-setting design that stored approved commands in plaintext in a file that could reach public repositories.
D inferred by us from the described behaviour, reason shown

Impact

Records exposed
2,539
C one secondary source states it
Other parties affected
63
C one secondary source states it
Severity
major
D inferred by us from the described behaviour, reason shown
Kind of harm
privacy, others' security
D inferred by us from the described behaviour, reason shown
Who was affected
1,122 public configuration files across 1,994 repositories; 63 organisations, primarily software companies; 2,539 secrets including database passwords, SSH credentials, GitHub/GitLab/npm tokens and AWS keys
C one secondary source states it
Kinds of data exposed
credentials, database passwords, ssh credentials, api tokens, aws keys
C one secondary source states it
Can it be undone
partly
D inferred by us from the described behaviour, reason shown
Why this severity
Credentials of 63 organisations were exposed in public GitHub repositories, touching many third parties.
D inferred by us from the described behaviour, reason shown

Context

Industry
Professional, Scientific, and Technical Services
D inferred by us from the described behaviour, reason shown
Shared exposure
one shared model or part put many users at risk
D inferred by us from the described behaviour, reason shown

LR-2026-0863 · version 16 · published 2026-11-15 after automatic checks; not yet checked field by field by a person