AWS released Trust in Depth, a framework for AI agents, on April 2024, saying it enables more autonomy without losing control. It is the company's first safety update since the introduction of Bedrock AgentCore in 2023.
AWS reported automated reasoning capabilities, which can verify agent behavior across all possible cases, measured on complex system interactions. That compares with earlier verification methods that only checked predefined scenarios.
Trust in Depth is built on AWS Bedrock and targets enterprise use cases requiring high levels of security and reliability. Availability begins in Q2 2024, initially for enterprise customers.
"Trust in Depth expands what you can safely allow by using independent layers that reinforce one another," said Anand Chandrasekaran, Senior Vice President of AWS. Automated reasoning checks whether rules hold in every possible case, ensuring agents behave as intended.
The announcement follows AWS's focus on AI governance and security. AWS said Trust in Depth aligns with its Defense in Depth strategy, enhancing control without limiting agent capabilities.
AWS did not say how Trust in Depth will handle ambiguous rules, and raised the question of how to balance autonomy with oversight. The company said it will continue refining the framework based on feedback.
Source: awsml
