AI helps find real cloud security threats across major providers

Validating AI-Discovered Cloud Threats on Live Infrastructure: A Multi-Model, Multi-Cloud Study

Cryptography and Security

Summary

Cloud security teams want better ways to find real threats in complex systems. The authors tested AI tools that suggest cloud security weaknesses, then tried out these threats safely on live cloud setups like AWS and Google Cloud. They confirmed about 75% of AI-suggested threats and found that using several AI models together finds more real issues than using just one. This shows AI can assist security checks when combined with actual tests on real cloud systems.

What this means in practice

  • For cloud security teams: Integrate multi-model AI threat discovery with live cloud testing to prioritize real security vulnerabilities effectively.
  • For security automation engineers: Build automated workflows that generate and validate exploit scenarios on live cloud systems for robust threat confirmation.

Authors

Nicholas Pecka, Lotfi Ben Othmane, Renee Bryce

Abstract

Cloud infrastructure threat modeling lacks empirical validation of AI-discovered threats. We present a three-phase methodology of multi-agent threat discovery, automated exploit synthesis, and non-destructive live validation, and apply it to three commercial foundation models on two cloud architectures, validating each threat natively on the cloud (AWS, GCP, or Azure) it was discovered against. Of 233 AI-discovered threats, 103 were confirmed and 34 refuted (75.2\% precision, 95\% CI [67.3, 81.7]); 44 were conditional (the named mechanism refuted but an equivalent impact shown through another path) and 52 indeterminate. Multi-model discovery helped through breadth, not agreement: the three models together found 2.2$\times$ the confirmed threats of the best single model at comparable precision, while threats reported by several models were confirmed no more often than single-model ones. False positives were not architecture-specific: they asserted access or system state the deployment did not have. Threats coded in advance as observable configuration resolved cleanly more often than multi-step behavioral threats (72.5\% vs.\ 51.6\%), which were more often conditional, suggesting that models more often misidentify the exploitation mechanism than the underlying weakness. AI-assisted discovery is a practical input to security workflows when paired with empirical validation.