Akamai Agent Sandbox and Inference Beta: Run Agent Code Isolated

The CDN company wants to run your AI agents. On October 27, Akamai launched two betas: Akamai Inference, a managed service for open-weight models, and Akamai Agent Sandbox, an isolated runtime for code that AI agents write. The sandbox is the interesting half. Agent-written code is untrusted code, and most small teams still run it on the same box as everything else.

What actually happened

Per Akamai's press release:

• Akamai Inference gives managed access to open-weight and custom models through OpenAI-compatible APIs, with serverless pay-per-use or dedicated deployments.
• It launches with 12 open-weight models from 8 billion to 2.8 trillion parameters, including NVIDIA Nemotron 3 Ultra, OpenAI gpt-oss-120b and Google Gemma 4, plus embedding models for retrieval.
• Agent Sandbox runs agents on Akamai Functions and executes the code they generate in a separate environment. It is an open-source component of the Spin framework and uses WebAssembly isolation to limit compute, memory, outbound connections and system capabilities.
• Akamai says it plans to add controls over what agents can reach and what can reach them. That is future tense.
• Both products are beta. The release publishes no product pricing.

Why the Akamai Agent Sandbox matters for your business

We have written about agent sandbox escapes more times than we would like. The pattern is the same each time: the model writes code, the code runs with the host's permissions, and the host has your keys. A WebAssembly sandbox flips the default. The code gets only the capabilities you hand it.

Three things we would do with this news:

Separate where the agent thinks from where its code runs. Even if you never touch Akamai, that split is the right architecture. Spin is open source, so you can test the isolation model locally before you commit to anyone's cloud.

Use the OpenAI-compatible API as an exit, not an entrance. If your app already talks to an OpenAI-shaped endpoint, pointing it at gpt-oss-120b or Gemma 4 on Akamai is a config change. Run your real prompts against it and compare cost and quality. Keep the endpoint in config, not in code.

Don't plan a budget around a beta with no price. No rates are public. Prototype on it. Don't move production until you have a number in writing.

Sources: Akamai: Akamai launches AI inference service and agent execution environment.