Threat reportThreat IntelligenceTL-2026-6154
Leaked Credentials on Public Repositories: GitGuardian Enterprise Good Samaritan Program and Agentic-AI (ARTEX) Abuse of Exposed Secrets
GitGuardian announced an Enterprise Good Samaritan Program that traces high-impact leaked credentials on public sources (GitHub, Docker Hub) to affected companies, citing 3,913+ high-impact leaks tied to 1,532 companies and only 35% of contacted organizations responding. Microsoft Threat Intelligence separately reported (2026-11-05) threat actors using the agentic AI framework ARTEX to mass-search GitHub for secrets and abuse them against Entra ID/Azure tenants, in one case going from GitHub search to Key Vault exfiltration in about five minutes. First disclosed .
Key facts
| Identifiers | TL-2026-6154 |
|---|---|
| Also known as | GitGuardian Enterprise Good Samaritan Program, ARTEX-driven secret abuse, Private-CISA GitHub leak |
| Severity | MEDIUM |
| First disclosed | |
| Affected products | Microsoft Entra ID service principals / Azure (Key Vault, Storage, App Service, AKS, AI services) with leaked application credentials; GitHub Public repositories containing hardcoded secrets |
| Weaknesses (CWE) | CWE-798, CWE-540 |
| Tools | ARTEX, Playwright headless Chromium |
| MITRE ATT&CK techniques | T1078.004 Cloud Accounts, T1090.003 Multi-hop Proxy, T1114.002 Remote Email Collection, T1190 Exploit Public-Facing Application, T1213.002 Sharepoint, T1526 Cloud Service Discovery, T1528 Steal Application Access Token, T1530 Data from Cloud Storage, T1552.001 Credentials In Files, T1552.004 Private Keys, T1555.006 Cloud Secrets Management Stores, T1580 Cloud Infrastructure Discovery and 3 more in the attack flow |
| Motivation | UNKNOWN |
| Target sectors | government administration, technology, managed-service-providers, cloud |
| Target regions | Global |
| Detection rules | 9 (1 shown free) |
| Sources | 8 cited sources |
Detection coverage
As of 2026-11-16, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-6154 across Splunk SPL, Microsoft KQL and Sigma, covering 8 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.
Free detection: TL-2026-6154 Leaked Secret Abuse - Entra Service Principal Failure-Then-Success and Multi-SPN Source IP
Behaviour-first detection of leaked application credentials being abused (ARTEX / Private-CISA style exposure). Flags a single source IP that hits Entra service principal sign-ins with invalid/expired secret errors (AADSTS7000215/AADSTS7000222) and then succeeds, or authenticates to 3+ service principals, plus any contact from research IOC IPs such as 199.19.77.84 (earliest seen 2026-10-28).
- Languages: Splunk SPL, Microsoft KQL, Sigma
- ATT&CK: T1078.004, T1528
- Data sources: azure_entra_serviceprincipal_signin_logs, azure:monitor:aad
- Severity: high
- Confidence: high
Splunk SPL
index=azure sourcetype="azure:monitor:aad" category="ServicePrincipalSignInLogs" earliest=-24h
``` TL-2026-6154: leaked app secret probing - AADSTS7000215/7000222 failures then success, one IP many service principals ```
| eval spn_id=coalesce('properties.servicePrincipalId', servicePrincipalId), src_ip=coalesce('properties.ipAddress', callerIpAddress), err=tonumber(coalesce('properties.status.errorCode', resultType, "0"))
| eval is_secret_fail=if(in(err,7000215,7000222),1,0), is_success=if(err=0,1,0)
| stats sum(is_secret_fail) as secret_failures sum(is_success) as successes dc(spn_id) as spn_count values(spn_id) as service_principals min(_time) as first_seen max(_time) as last_seen by src_ip
| eval known_ioc=if(in(src_ip,"199.19.77.84","199.19.77.18","65.87.223.26","157.22.126.128"),1,0)
| where known_ioc=1 OR (spn_count>=3 AND secret_failures>=1 AND successes>=1)
| convert ctime(first_seen) ctime(last_seen)
| table src_ip spn_count secret_failures successes known_ioc service_principals first_seen last_seenMicrosoft KQL
// TL-2026-6154 - Entra service principal sign-ins with leaked application credentials
let iocIps = dynamic(["199.19.77.84","199.19.77.18","65.87.223.26","157.22.126.128"]);
AADServicePrincipalSignInLogs
| where TimeGenerated > ago(24h)
| extend Err = toint(ResultType)
| summarize SecretFailures = countif(Err in (7000215, 7000222)), Successes = countif(Err == 0),
SpnCount = dcount(ServicePrincipalId), ServicePrincipals = make_set(ServicePrincipalName, 25),
FirstSeen = min(TimeGenerated), LastSeen = max(TimeGenerated) by IPAddress
| where IPAddress in (iocIps) or (SpnCount >= 3 and SecretFailures >= 1 and Successes >= 1)
| project IPAddress, SpnCount, SecretFailures, Successes, ServicePrincipals, FirstSeen, LastSeenSigma
title: Entra Service Principal Sign-In Failure - Invalid or Expired Client Secret (TL-2026-6154)
id: 6f1a2c3e-4b5d-4e6f-8a7b-9c0d1e2f3a41
status: experimental
description: Detects Entra ID service principal sign-in failures AADSTS7000215 (invalid client secret) and AADSTS7000222 (expired client secret). Microsoft reports leaked-secret abuse by ARTEX-driven actors surfaces as these failures followed by a success from the same IP. Correlate per IP and app with later successful sign-ins.
references:
- https://www.microsoft.com/en-us/security/blog/2026/11/05/threat-actors-use-agentic-ai-tooling-to-find-and-abuse-exposed-secrets/
- https://blog.gitguardian.com/how-we-got-a-cisa-github-leak-taken-down-in-26-hours/
author: Threadlinqs Team
date: 2026/11/16
tags:
- attack.initial-access
- attack.t1078.004
- attack.credential-access
- attack.t1528
logsource:
product: azure
service: signinlogs
detection:
selection:
ResultType:
- '7000215'
- '7000222'
condition: selection
falsepositives:
- Application owners with stale or misconfigured client secrets after rotation
- Automation or CI jobs still using a retired secret
level: medium
Known false positives:
- Developers testing a rotated or expired secret from a corporate or CI egress IP
- Multi-tenant SaaS or MSP automation legitimately authenticating to several service principals from one IP
Tuning notes: Replace index/sourcetype with your Azure Monitor AAD input. Allow-list corporate/CI egress IPs and known automation SPNs; raise spn_count threshold for MSP tenants; keep the IOC list in sync with research.json.
- TL-2026-6154 App Service Exploitation - Shell Commands in User-Agent and Auth Token Probing · unlock with TL Creds
- TL-2026-6154 Post-Compromise - Key Vault Secret Enumeration Burst by Application Identity · unlock with TL Creds
- TL-2026-6154 C2 / Proxy Infrastructure - Azure Activity From ARTEX-Linked IPs and Tor Exit · unlock with TL Creds
- TL-2026-6154 Credential Exposure - GitHub Repo Made Public or Secret Scanning Weakened · unlock with TL Creds
- TL-2026-6154 Cloud Administration Command - Azure VM Run Command by Service Principal · unlock with TL Creds
- TL-2026-6154 ARTEX Tooling - Headless Browser and Nmap Fingerprints Against Web Apps · unlock with TL Creds
- TL-2026-6154 Post-Compromise - ARM Secret Harvesting by Service Principal (Storage Keys, App Settings, AKS Credentials) · unlock with TL Creds
- TL-2026-6154 Collection - App-Only Microsoft Graph Mail and SharePoint/OneDrive Reads by Service Principal · unlock with TL Creds
Indicators of compromise
As of 2026-11-16, Threadlinqs Intelligence tracks 8 indicators of compromise for TL-2026-6154.
Sample indicators (3 of 8)
- IP address:
199.19.77[.]84 - IP address:
199.19.77[.]18 - IP address:
65.87.223[.]26
Defanged, TLP:CLEAR. Full set of 8 indicators: unlock with TL Creds.
Technical analysis
Credential exposure on public code platforms remains a high-volume initial-access enabler. GitGuardian's 2026 State of Secrets Sprawl report counts about 28.65 million new hardcoded secrets on public GitHub in 2025 (+34% year over year), and the vendor states roughly 90% of valid secrets remain unrevoked one week after the leaker is alerted. On 2026-11-16 GitGuardian announced an Enterprise Good Samaritan Program (extending its developer-focused program running since 2018) that combines AI-driven attribution with human review to identify which organization a leaked credential affects, then discloses to multiple people at that organization (including security teams), regardless of customer status. Vendor-reported figures: 3,913+ high-impact leaks linked to 1,532 companies, follow-up reaching at least five people per organization, and a 35% response rate. These figures come from a vendor press release and were not independently verified.
A May 2026 case study illustrates the exposure blast radius: a public GitHub repository named 'Private-CISA' (844 MB; created 2025-11-13) contained CI/CD logs, Kubernetes manifests, ArgoCD files, Terraform, GitHub Actions workflows, internal documentation exports, plain-text passwords, AWS tokens, Entra ID SAML certificates, private keys, GitHub tokens, and instructions for disabling GitHub secret scanning; some credentials were still valid. Nine automated Good Samaritan emails to the commit author went unresolved; after reporting to CERT/CC on 2026-05-14 and escalating through personal contacts, the repository was taken offline on 2026-05-15, about 26 hours after detection.
Microsoft Threat Intelligence (blog dated 2026-11-05) documents the offensive side: since August 2026 it observed large volumes of GitHub secret-search queries combining a target organization name with secret keywords, cloud-service keywords, or AI-provider keywords. Stolen still-valid application credentials were then used for service principal sign-ins to Entra ID tenants (earliest observed 2026-10-28 from 199.19.77.84), followed by discovery of subscriptions and resources, retrieval of storage account keys, App Service settings, connection strings and AKS credentials, ARM-based enumeration and exfiltration of Key Vault secrets, OAuth token theft for single-tenant apps, Azure DevOps secret searching, Azure VM Run Command execution, and SharePoint/OneDrive/mailbox collection via Microsoft Graph. The agentic framework ARTEX adapted failed ARM calls within roughly 28-80 seconds (e.g., correcting the API version to 2023-07-01). Microsoft did not name an actor for the GitHub scanning or the Entra/Azure activity; secondary coverage notes Storm-0940 previously used ARTEX and that CrowdStrike reported ARTEX-supported espionage against 200+ organizations since mid-October (secondhand, not verified here). BeaconBeagle lookups for the network IOCs returned HTTP 404 (no data available).
MITRE ATT&CK techniques by tactic:
Initial Access
T1078.004 Cloud Accounts; T1190 Exploit Public-Facing Application
Command and Control
Collection
T1114.002 Remote Email Collection; T1213.002 Sharepoint; T1530 Data from Cloud Storage
Discovery
T1526 Cloud Service Discovery; T1580 Cloud Infrastructure Discovery
Credential Access
T1528 Steal Application Access Token; T1552.001 Credentials In Files; T1552.004 Private Keys; T1555.006 Cloud Secrets Management Stores
Reconnaissance
T1589.001 Credentials; T1593.003 Code Repositories
Execution
Remediation
Immediate actions
- Treat any credential exposed in a public repository as compromised, even if the repository or commit was later deleted; revoke and rotate it
- Rotate everything stored in any Key Vault that an affected application or service principal could read
- Remove all client secrets, certificates and federated credentials from affected Entra apps (including attacker-added ones); revoke Entra sessions, Azure DevOps PATs and SSH keys
- Block 199.19.77.84/32 with a Conditional Access named location for workload identities (requires Workload ID Premium; single-tenant apps only)
- Hunt Entra service principal sign-in logs for AADSTS7000215/AADSTS7000222 failures followed by success, and one IP authenticating to 3+ service principals
Workarounds
- Restrict who can register apps and limit the Application Developer role
- Apply Conditional Access and workload identity risk protection
Longer-term hardening
- Enable GitHub secret scanning with push protection and GitHub Advanced Security for Azure DevOps
- Audit the organization's public GitHub footprint, including contractor and developer personal repositories
- Publish a security.txt file and staff a monitored security inbox for external disclosures
- Migrate OAuth apps and service principals to managed identities; enforce least-privilege RBAC and app management policies capping secret lifetimes
- Enable Defender for Cloud plans (Resource Manager, Storage, Key Vault, App Service, Databases) and consider honeytokens/deception controls
Affected and fixed versions
- Microsoft — Entra ID service principals / Azure (Key Vault, Storage, App Service, AKS, AI services) with leaked application credentials
Vulnerable versions: Any tenant with exposed application secrets - GitHub — Public repositories containing hardcoded secrets
Vulnerable versions: N/A - configuration/hygiene exposure
Timeline
- Public GitHub repository 'Private-CISA' (844 MB) is created, exposing CI/CD logs, Kubernetes/ArgoCD/Terraform files, plain-text passwords, AWS tokens, Entra ID SAML certificates, private keys and GitHub tokens (per GitGuardian).
- GitGuardian publishes the State of Secrets Sprawl 2026: 28.65 million new secrets on public GitHub in 2025 (+34% YoY).
- GitGuardian's Good Samaritan program had sent nine automated emails to the Private-CISA commit author without resolution.
- GitGuardian reports the Private-CISA leak to CERT/CC at 4:14 PM CET; only an automated acknowledgment follows.
- After escalation through personal contacts, GitGuardian reaches CISA and the repository is taken offline roughly 26 hours after detection.
- Microsoft begins observing large volumes of GitHub secret-search queries tied to agentic AI tooling (reported as 'August 2026').
- Microsoft observes the earliest service principal sign-ins from locked-ioc-2 using leaked application credentials against downstream Entra ID tenants.
- Microsoft Threat Intelligence publishes its report on threat actors using agentic AI tooling (ARTEX) to find and abuse exposed secrets, including a ~5-minute GitHub-to-Key-Vault case.
- GitGuardian announces the Enterprise Good Samaritan Program: 3,913+ high-impact leaks tied to 1,532 companies, 35% response rate.
Sources cited
- GitGuardian Enterprise Good Samaritan Program gitguardian.com · Vendor report
- Microsoft Threat Intelligence: Threat actors use agentic AI tooling to find and abuse exposed secrets Microsoft · Vendor report · 2026-11-05
- The State of Secrets Sprawl 2026 blog.gitguardian.com · Research · 2026-03-17
- CSA Research Note: Five Minutes from Search to Key Vault - Agentic Secret Harvesting labs.cloudsecurityalliance.org · Research · 2026-11-09
- How we got a CISA GitHub leak taken down in 26 hours blog.gitguardian.com · Blog
- GitGuardian: Why we built the Good Samaritan Program blog.gitguardian.com · Blog
- Leaked Credentials Are an Enterprise Problem: GitGuardian Extends Its Good Samaritan Program to Security Teams cybersecuritynews.com · News · 2026-11-16
- Cybersecurity Dive: Hackers use ARTEX to find exposed secrets in GitHub repositories cybersecuritydive.com · News · 2026-11-06
Previous and next in threat intel
Drafted by the Threadlinqs research pipeline from 8 sources; published ; last updated —.