Deno Land, maker of the Deno runtime, has introduced Deno Sandbox, a secure environment built for code generated by AI agents. The company also announced the long-awaited general availability of Deno Deploy, a serverless platform for running JavaScript and TypeScript applications. Both were announced on February 3.
Now in beta, Deno Sandbox offers lightweight Linux microVMs running as protected environments in the Deno Deploy cloud. Deno Sandbox defends against prompt injection attacks, the company said, where a user or AI attempts to run malicious code. Secrets such as API keys never enter the sandbox and will only appear when an outbound HTTP request is sent to a pre-approved host, according to the company.
Deno Sandbox was created in response to the rise in AI-driven development, explained Deno co-creator Ryan Dahl, as more LLM-generated code is being released with the ability to call external APIs using real credentials, without human review. In this scenario, he wrote, “Sandboxing the compute isn’t enough. You need to control network egress and protect secrets from exfiltration.” Deno Sandbox provides both, according to Dahl. It specializes in workloads where code must be generated, evaluated, or safely executed on behalf of an untrusted user.
