How Benchling secured multi-tenant AI agents with Amazon Bedrock AgentCore
aws ·
Benchling successfully implemented a secure, multi-tenant AI agent architecture using Amazon Bedrock AgentCore to execute scientific code for thousands of life sciences tenants. To address the limitations of traditional sandboxing, particularly regarding DNS-based data exfiltration, they deployed a defense-in-depth strategy within an isolated AWS account. The solution utilizes Amazon Route 53 Resolver DNS Firewall with a strict deny-list, allow-list, and catch-all policy to seal network boundaries. VPC endpoint policies further restrict access to only specific S3 buckets, ensuring that untrusted code cannot reach unauthorized resources even if credentials are compromised. Benchling avoids unsustainable IAM role sprawl by injecting per-job scoped credentials via AWS Security Token Service (STS) at runtime. This approach has processed over 600 execution sessions daily across more than 250 tenants weekly with zero security incidents or data leakage. By leveraging managed infrastructure instead of building custom sandboxes, Benchling maintains rigorous security controls while focusing on core product development.
aws