Hackers turn to agentic AI to find exposed secrets on GitHub, warns Microsoft
Hackers are using AI tools to find exposed passwords and API keys on GitHub.
Microsoft found that attackers used AI to access cloud systems and steal sensitive data.
Microsoft advises organisations to secure their credentials, monitor suspicious activity and strengthen security.
Microsoft has recently warned the users across the world, as their Threat Intelligence unit found that hackers are using AI to find and abuse exposed secrets. The threat actors are using agentic tools to find exposed passwords, application credentials and API keys on public GitHub repositories and use them to break into cloud systems. Not only that, but the company also claims that these automated tools can scan multiple repositories, identify valuable credentials and carry out follow-up attacks with limited human involvement. The company also shared an incident where they found that the AI agent moved from searching GitHub for exposed secrets to stealing data from an organisation's Azure Key Vault in just five minutes.
AI agents turn exposed GitHub secrets into security risks
Microsoft shared a blog post recently where they said that the malicious activity began in August 2026. They pointed out that the attackers used an agentic AI framework called ARTEX to search public GitHub repositories for sensitive information. While investigating, Microsoft detected successful automated login attempts from the IP address 199.19.77[.]84 using multiple application IDs across several affected customers.
Not only that, but in some cases the attackers had stolen the credentials of a vendor-owned application that was used by multiple organisations, allowing them to access several customer environments. The company also shared that the AI tool searched cloud resources, accessed Azure Key Vaults and Storage Accounts, downloaded files from SharePoint and OneDrive, and attempted to obtain additional credentials. Furthermore, some compromised applications were also used to explore AI and machine learning services.
In certain cases, attackers targeted credentials belonging to multitenant applications, potentially allowing access to several customer environments through a single compromised application.
What did Microsoft say?
Microsoft has advised organisations to treat every publicly exposed credential as compromised, even if the original code has been edited or deleted. Do note that removing exposed information does not invalidate the credentials.
The company also added that the organisations should immediately revoke or rotate affected passwords, API keys, application secrets and certificates. Microsoft also recommended reviewing Entra ID sign-in records, Azure activity logs and Key Vault access records for suspicious behaviour.
How to prevent AI-powered attacks
Microsoft recommends that organisations take steps to protect themselves from AI-powered cyberattacks. They should avoid storing long-term passwords or secret keys in their code and use managed identities instead.
They should also give users and applications only the access they need, enable GitHub tools to detect and block exposed secrets, and improve security monitoring using Microsoft Defender for Cloud.
Microsoft also recommends using honeypots and decoy systems. These are fake systems designed to attract attackers, helping security teams detect suspicious activities and identify potential threats.