Microsoft Threat Intelligence said on November 5, 2026 that attackers use agentic AI, software that plans and acts alone, to hunt for leaked secrets in public GitHub code. Since August 2026, it saw many searches naming a target company plus a secrets keyword. In one case, an agent went from a GitHub search to taking Azure Key Vault data in five minutes.
Microsoft describes the chain. A secret sits in public code. The agent finds it and signs in as the app. It lists cloud resources, pulls credentials from Key Vault, then downloads files. When a request failed, it changed the method or version and retried. Separately, as early as October 28, 2026, Microsoft saw attacks using leaked or stolen app credentials. It does not say each one came from GitHub. In some cases a vendor's multitenant app was compromised and used to reach other companies' cloud accounts.
A secret in code was once found only when a person looked. Now the finder can act within minutes. Microsoft says exposed credentials stay usable until revoked or rotated, and deleting the code does not fix that. Leaders can ask which apps still use stored secrets, how long an exposed key stays valid, and who owns rotation. Managed identities store no secret, so they break the first link.