How Benchling secured multi-tenant AI agents with Amazon Bedrock AgentCore
Benchling leveraged Amazon Bedrock AgentCore’s Code Interpreter in VPC mode to securely run AI‑generated scientific code for thousands of life‑science tenants, processing over 600 execution sessions per day across more than 250 tenants per week with zero security incidents. The design combines account‑level isolation, Amazon Route 53 Resolver DNS Firewall, and VPC endpoint policies to enforce strict per‑job data access and block all unauthorized network traffic, including DNS resolution. This defense‑in‑depth architecture lets engineers scale multi‑tenant AI agents without proliferating IAM roles while maintaining full control over network isolation.
⚡ Key Takeaways
- Processes >600 code execution sessions per day across >250 tenants per week with zero security incidents.
- Uses Amazon Bedrock AgentCore Code Interpreter in Amazon VPC mode.
- Enforces network isolation via Amazon Route 53 Resolver DNS Firewall and VPC endpoint policies.
- Avoids per‑tenant IAM role sprawl by centralizing security controls in the Untrusted Code Account.
- Requires no outbound HTTP/Egress ports and blocks DNS resolution unless explicitly allowed.
Engineers can now deploy multi‑tenant AI agents that execute untrusted code at scale while guaranteeing tenant data isolation and preventing exfiltration, all without the operational overhead of managing thousands of IAM roles.
✅ Practical Steps
- Deploy the AgentCore Code Interpreter in Amazon VPC mode as described, routing execution traffic through a dedicated Untrusted Code Account.
- Configure Amazon Route 53 Resolver DNS Firewall rules to whitelist only approved domains for each execution job.
- Apply VPC endpoint policies that restrict reachable AWS service endpoints to the minimum required for the tenant’s workload.
Want the full story? Read the original article.
Read on AWS ML Blog ↗