Abnormal AI has put Amazon Bedrock AgentCore’s Code Interpreter into production as the compute layer behind its email threat detection agents. The zero trust sandbox claim behind it deserves a closer look given what independent researchers found in the same feature last year.