We Found Eight Attack Vectors Inside AWS Bedrock. Here's What Attackers Can Do with Them

XM Cyber research analyzes AWS Bedrock and documents eight validated attack vectors — including model invocation log tampering, knowledge‑base data‑source and data‑store compromise, direct and indirect agent hijacking, flow injection, guardrail weakening, and managed prompt poisoning — showing how over‑privileged identities and misconfigured integrations can enable data exfiltration, credential theft, lateral movement, and large‑scale prompt compromise across AI workflows.