AI Systems Can Independently Compromise Cloud Security with Limited Supervision

Post Views: 5

Palo Alto Networks Develops AI System to Test Cloud Security

Researchers from Palo Alto Networks’ Unit 42 team have created an artificial intelligence system called Zealot, which can autonomously execute sophisticated attacks on cloud infrastructure.

According to the researchers, “Zealot was designed to empirically test the capabilities of AI systems against live cloud environments.”

How Zealot Works

The AI system operates under a ‘supervisor-agent’ model, delegating tasks to three specialized sub-agents:
One for infrastructure reconnaissance and network mapping,
one for web application exploitation and credential extraction,
and one for cloud security operations.

Results and Observations

Without additional guidance, Zealot autonomously scanned the network, identified a connected virtual machine, exploited a web application vulnerability to obtain credentials, and eventually extracted the target data, even gaining additional permissions when encountering access barriers.

However, the researchers observed instances of unproductive loops, where the AI fixated on irrelevant targets and wasted resources until human operators intervened.

Implications and Recommendations

Existing detection systems, built around human attacker behavior patterns, are ill-suited to identify AI-driven intrusions that move rapidly and leave distinct digital footprints.

Therefore, researchers recommend that organizations proactively audit cloud permissions, restrict access to metadata services, and adopt AI-powered defenses to counter the growing threat posed by AI-driven attacks.