Autonomous AI Agents Compromise Thousands of Credentials in Under Six Hours

Severity
Medium

Detail

Google Threat Intelligence Group (GTIG) has reported a growing trend of threat actors using AI and autonomous agents to accelerate cyberattacks, including credential theft, vulnerability discovery, malware development, and cloud resource abuse. One financially motivated threat actor, TeamPCP, used an autonomous multi-agent framework to conduct a large-scale credential harvesting campaign, compromising thousands of third-party credentials in less than six hours.

The attackers are also increasingly targeting AI coding assistants, developer environments, open-source repositories, and enterprise AI infrastructure. Stolen credentials and access are subsequently monetized through direct sales, ransomware operations, and data extortion.

Google also observed threat actors compromising cloud environments to deploy local AI models, steal proprietary AI research and models, and use generative AI to assist with reconnaissance, social engineering, exploit development, and malware creation.

How?

The attack involved compromising cloud infrastructure and using an AI coding chatbot together with predefined prompts and agent instructions to automate the credential harvesting process.

The autonomous agents managed vulnerability scanning, identified targets, performed credential harvesting, troubleshot issues in real time, and rotated IP addresses without requiring continuous human intervention.

TeamPCP has also used credential-stealing malware such as SANDCLOCK and DUSTMAKER. DUSTMAKER is designed to target CI/CD environments and steal cloud, developer, and other sensitive credentials. Attackers have additionally experimented with prompt injection and poisoned AI assistant workspaces to influence AI tools and evade security controls.

Conclusion

The increased use of autonomous AI agents demonstrates how attackers can perform traditionally time-consuming activities at significantly greater speed and scale. AI is increasingly being used not only to generate malicious code but also to automate reconnaissance, credential theft, vulnerability research, social engineering, and attack execution.

Security teams should closely monitor AI development environments, CI/CD pipelines, cloud credentials, API keys, and unusual AI-agent activity. Organizations should also protect developer environments, enforce strong authentication, rotate exposed credentials, restrict cloud permissions, and monitor for unauthorized use of AI workloads.

Finally, organizations should establish security controls around AI coding assistants and locally deployed AI models, particularly where sensitive source code, credentials, or proprietary data may be exposed.

Source

https://thehackernews.com/2026/09/autonomous-ai-agents-compromise.html