AWS Amazon Bedrock AgentCore security and payment updates

Amazon Web Services (AWS) has introduced new technical capabilities through Amazon Bedrock AgentCore to manage security, authentication, and payments for AI agents.

Initial implementations focused on authentication and financial controls. To support legacy enterprise environments, organizations can use a Lambda interceptor to enable HTTP Basic Authentication via AWS Secrets Manager, though AWS recommends modernizing to OAuth 2.0, SAML, or IAM to mitigate risks like prompt injection. For financial transactions, the ‘aws-agents-pay’ plugin allows OpenClaw agents to perform bounded HTTP 402 payments. This system enables operators to set per-payment ceilings and cumulative budgets to prevent agents from exceeding spending limits.

Subsequent developments addressed data access and downstream authorization. To prevent unauthorized information retrieval, AWS is shifting authorization responsibility from the AI agent to downstream services. In this model, the agent serves as an orchestrator while user identity and department-specific permissions are carried through authentication tokens from providers such as Amazon Cognito, Microsoft Entra ID, or Okta. This ensures underlying services enforce access controls based on actual user credentials rather than the agent itself.