The scenario that has lived on CISO briefing slides for two years arrived in production this week. Microsoft documented threat actors using autonomous AI agents to find exposed OAuth secrets on GitHub and compromise tenants - no human in the loop between discovery and exploitation. That alone would make for a full issue. But the identity news did not stop there: Chainalysis escalated its breach to a full credential dump, DigiCert exposed customer passports through a migration error, and Google put a dollar figure on LLMjacking that will get CFO attention.

Every story in today’s brief traces back to the same root cause: non-human identities and credentials that are poorly inventoried, poorly rotated, and poorly monitored.

In the News

AI Agents Autonomously Mine GitHub for OAuth Secrets and Compromise Tenants

Microsoft’s threat intelligence team published findings documenting the first confirmed case of AI agents - referred to as ARTEX tooling - being deployed by threat actors to autonomously search public GitHub repositories for exposed OAuth application secrets. The agents identify valid client IDs and secrets, authenticate into production Entra ID tenants, and begin reconnaissance, all without human intervention at the exploitation stage.

The operational significance is the compression of the exposure-to-compromise window. Traditional secret exposure on GitHub gave defenders a window - sometimes hours, sometimes days - between a credential appearing in a public commit and an attacker manually discovering and using it. ARTEX collapses that window to minutes. Microsoft observed tenant compromise occurring faster than most secret-scanning tools can alert.

This is the practical arrival of the “AI finds the credential, AI exploits the credential” attack chain. The defensive response is not new - secrets management, automated rotation, OAuth app inventory, and anomaly detection on service principal sign-ins are all established controls. What changed is the urgency. Manual rotation cadences measured in days or weeks are now functionally equivalent to no rotation at all.

What defenders should do: Inventory all OAuth app registrations in Entra ID. Enable automated secret rotation with a cadence under 24 hours for internet-facing app credentials. Deploy identity threat detection and response (ITDR) tooling that flags service principal sign-ins from unexpected locations or at unexpected volumes. Run TruffleHog or GitHub Advanced Security scanning on every repository.

Chainalysis Breach Escalates to Full Credential Dump

The Chainalysis breach first disclosed on November 3 - initially described as an agentic AI attack targeting internal systems - has escalated significantly. The company now confirms that all user credentials, password hashes, and live session tokens were compromised. Reactor and Data Solutions platforms remain offline as of this writing.

The downstream impact is substantial. Chainalysis is the primary blockchain analytics platform used by law enforcement agencies, financial regulators, and compliance teams for crypto-fraud investigations. A full credential dump means every integrated system that accepted Chainalysis credentials or session tokens is now potentially exposed. Customers face credential-reset obligations not just for Chainalysis itself, but for any system where those credentials were reused.

What defenders should do: Force credential resets for any accounts associated with Chainalysis. Revoke and reissue all API tokens and session keys. Deploy phishing-resistant passwordless authentication (FIDO2) rather than simply resetting passwords - if the hashes are in the wild, a password reset only buys time until the new hash is cracked. Monitor for session token replay from anomalous devices or geolocations.

DigiCert Breach Exposes Customer Identity Documents

DigiCert, one of the largest certificate authorities in the PKI ecosystem, disclosed a data breach caused by a server migration misconfiguration that left an internal business system exposed to the internet. The compromised data includes customer identity verification documents - passports, driver’s licenses, and other government-issued IDs submitted as part of the certificate validation process.

The breach is operationally significant beyond the PII exposure. Certificate authorities occupy a foundational trust position in internet security. Organizations submit sensitive identity documents to CAs precisely because they trust the CA’s security posture. A configuration error during a routine migration - the kind of mistake that external attack surface management and ZTNA enforcement are designed to prevent - erodes that trust at a systemic level.

DigiCert has not disclosed the number of affected customers or the duration of exposure, according to public reporting from BleepingComputer.

What defenders should do: Organizations that submitted identity documents to DigiCert for certificate validation should assume those documents are compromised and enable identity monitoring services. More broadly, this is a reminder that configuration management during infrastructure migrations - particularly for systems handling PII - requires pre-migration and post-migration external attack surface scans and ZTNA policies that default-deny external access to internal business systems.

LLMjacking Losses Hit $1M Per Enterprise

Google confirmed that LLMjacking - the theft of cloud AI API keys to run unauthorized LLM workloads on victim accounts - has caused losses exceeding $1M for individual enterprises. Stolen API keys are acquired through phishing campaigns, public code repository exposure, and cloud misconfigurations, then traded on Telegram markets and used within hours.

This is the non-human identity version of cryptojacking, with a more painful billing profile. A compromised API key with no usage caps or anomaly alerts can generate six-figure compute charges before anyone notices. The root cause is the same pattern driving the ARTEX story above: API keys and service credentials that are exposed, unrotated, and unmonitored.

What defenders should do: Implement budget alerts and usage anomaly detection on all cloud AI API keys. Set hard spend caps where the provider supports them. Rotate API keys on a cadence shorter than the attacker’s typical time-to-use (under 24 hours for high-value keys). Run secrets scanning in CI/CD pipelines to catch keys before they reach public repositories.

Defender Action Items

  • Inventory all OAuth app registrations and service principals in Entra ID; remove unused registrations immediately
  • Enable automated secret rotation with sub-24-hour cadence for internet-facing application credentials
  • Deploy ITDR with anomaly detection on service principal sign-ins - flag new geolocations, unusual volumes, off-hours authentication
  • Force credential resets and session token revocation for any Chainalysis-integrated accounts; deploy FIDO2 passwordless auth rather than password-only resets
  • Set hard budget caps and usage anomaly alerts on all cloud AI API keys; integrate billing anomalies into SIEM correlation
  • Run external attack surface scans before and after every infrastructure migration - validate that internal-only systems remain internal-only

Detection Queries

Service principal sign-ins from unexpected locations in Entra ID, surfaced via Splunk:

index=azure sourcetype="azure:aad:signin" properties.appId=* properties.servicePrincipalId=*
| stats count dc(properties.ipAddress) as unique_ips values(properties.ipAddress) as src_ips by properties.appDisplayName properties.servicePrincipalId
| where unique_ips > 3
| sort - unique_ips

This query identifies OAuth applications and service principals authenticating from more than three distinct IP addresses - a high-fidelity signal for credential compromise when the application is expected to authenticate from a known set of infrastructure IPs. Tune the threshold based on your environment. False positive rate is low for applications with static infrastructure; higher for globally distributed services.

References

Subscribe to The Identity Brief

Get The Identity Brief in your inbox (Mon/Wed/Fri) - Human, machine, and AI identity security — NHI, ITDR, and the IAM market.