Study confirms 103 of 233 AI-discovered cloud threats on live clouds
Original title: Validating AI-Discovered Cloud Threats on Live Infrastructure: A Multi-Model, Multi-Cloud Study
Three foundation models reached 75.2% precision, but multi-model agreement did not improve confirmation rates.
Researchers tested a three-phase method of multi-agent threat discovery, automated exploit synthesis, and non-destructive live validation using three commercial foundation models on AWS, GCP, and Azure. Of 233 AI-discovered threats, 103 were confirmed and 34 refuted, for 75.2% precision; 44 were conditional and 52 indeterminate. The three models together found 2.2 times the confirmed threats of the best single model, but threats reported by several models were not confirmed more often. Configuration threats resolved cleanly more often than multi-step behavioral threats.
Unconfirmed: The paper does not name the three commercial foundation models.
- Precision was 75.2%, with a 95% confidence interval of 67.3 to 81.7.
- Forty-four threats were conditional and 52 were indeterminate.
- Observable configuration threats resolved at 72.5%, versus 51.6% for behavioral threats.
- Each threat was validated on the cloud where it was discovered.
From the original article
1 min read at arxiv.orgCloud 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.
By Nicholas Pecka · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.