Shipping AI agents used to be a product question. Now it's a compliance question. The EU AI Act, ISO 42001, NIST AI RMF, and the OWASP LLM Top 10 have raised the bar for anyone running AI in production and the bar isn't "do you have controls," it's "can you prove it."

In this 45-minute session, AWS and DarkHunt split the problem along the shared responsibility line. AWS walks through how security requirements scale with the scope and autonomy of your AI and the Amazon Bedrock AgentCore platform that secures the model and runtime side: agent identity, authorization, credential brokering, session isolation, and observability. Darkhunt will then explain how this can be tested through their adversarial validation catalog and a live indirect-injection finding, with remediation that hardens both stacks together.

You'll leave with a five-step plan your team can start on this week.