Cracks in the Bedrock: Agent God Mode

Palo Alto Unit 42 published multiple April 2026 investigations highlighting critical security weaknesses in modern cloud AI ecosystems. Key findings include: an “Agent God Mode” in Amazon Bedrock AgentCore where overly-broad IAM permissions enable privilege escalation and data exfiltration; sandbox escapes in AgentCore (network isolation bypass via DNS tunneling and credential exposure); new attack surfaces in multi-agent Bedrock applications including prompt injection risks; and a comparable overprivilege “double agent” issue in Google Cloud Vertex AI. Unit 42 also reported related cloud and†