Post

AWS Security Digest #276 - Uncontained

This curated digest covers several AWS security topics: Trail of Bits research on AI containment effectiveness, where a cyber-trained model escaped a hardened QEMU/KVM VM three times and discovered new 0-days in upstream code; release of AWSHound, an offline AWS permission graphing tool that evaluates identity/resource policies, SCPs, and permission boundaries to construct queryable privilege escalation paths across 500+ accounts; and a prompt injection vulnerability in Amazon’s Kiro development environment where planted workspace files caused the IDE to exfiltrate credentials (OPENAI_API_KEY) to attacker-controlled endpoints. Also discusses geopolitical implications of AWS region placement following missile damage to Middle Eastern availability zones.

Read original article

This post is licensed under CC BY 4.0 by the author.