Hackers Now Use AI Agents to Find Your Secrets

Attackers are now using AI agents to automatically find and exploit exposed secrets on public code repositories like GitHub, Microsoft reports.

Attackers have a powerful new tool for finding and exploiting sensitive information. According to a new report from Microsoft Threat Intelligence, malicious actors are now using agentic AI frameworks to automate their attacks. These AI-powered tools actively scan public GitHub repositories to discover exposed secrets, such as API keys, database credentials, and other application secrets. Once found, the AI can then use these credentials to compromise cloud services. This marks a significant evolution from older, less sophisticated methods, representing a new and faster threat for any organization with code on public platforms. The observation confirms that attackers are weaponizing the same AI technologies that companies are adopting for productivity, turning them into highly efficient instruments for cybercrime.

What makes these agentic AI attacks different is their autonomy and intelligence. Previous methods for finding exposed secrets relied on simple scripts that used regular expressions to search for patterns that looked like keys. This approach was often noisy, generating many false positives and requiring a human to manually verify each potential finding. An AI agent, however, can operate with a high-level goal, such as “find and validate credentials for a cloud provider.” It can then independently devise and execute a series of steps to achieve this. This includes not just finding a potential secret, but also attempting to use it to authenticate against a service. This self-directed, multi-step process makes the attack far more efficient, allowing attackers to validate and exploit a vulnerability in a single, automated workflow at a scale and speed previously unimaginable.

This development is the latest move in the ongoing arms race between cybersecurity defenders and attackers. For years, security companies have been integrating machine learning and AI into their products to detect anomalies, identify new malware, and predict threats. Now, offensive actors are catching up by building their own AI-driven tools. The emergence of agentic AI for offensive purposes lowers the barrier to entry for sophisticated attacks. What once required a skilled team of hackers can now potentially be accomplished by a single individual using an automated framework. This democratization of advanced attack capabilities means that more organizations will face highly effective, automated threats, regardless of their size or industry. The cat-and-mouse game of cybersecurity is now increasingly becoming an AI-versus-AI battleground.

The practical takeaway for developers, CTOs, and security teams is a dramatic reduction in the time they have to respond to a mistake. The window between a developer accidentally committing a secret to a public repository and an attacker exploiting it has shrunk from hours or days to mere minutes. This reality makes reactive security measures, like manually scanning for leaks after the fact, dangerously inadequate. The focus must shift decisively toward prevention. This includes implementing automated secret scanning directly into CI/CD pipelines to block sensitive data before it ever becomes public. It also reinforces the critical importance of cloud security fundamentals, such as enforcing the principle of least privilege, so that even if a key is compromised, the potential damage is strictly limited. In this new era of AI-driven attacks, automated and proactive defense is no longer optional.

This development means the time from an accidental credential leak to a full-blown cloud compromise is now measured in minutes. AI agents can automate the entire attack chain from discovery to exploitation, operating at a scale and speed that manual security teams cannot match.

The use of AI agents by attackers significantly increases the risk and potential cost of a data breach. A single developer mistake can now lead to a rapid, automated compromise of company cloud infrastructure, threatening customer data, intellectual property, and business continuity.