How Sublime's AI agents are secure by design · Blog · Sublime Security

This post describes Sublime’s security-first architecture for two email security AI agents (ASA and ADÉ), emphasizing narrow scope, platform-enforced access controls, human-in-the-loop defaults, transparency with evidence-backed reasoning, protections against prompt injection, and data residency on AWS Bedrock within the customer environment to minimize risk and enable auditable, recoverable operations.