Akamai Inference Agent Sandbox Gives AI Agents a Safe Cage

Akamai introduced Akamai Inference Agent Sandbox on October 27, 2026, adding managed model serving and isolated execution for AI agents to Akamai Cloud. The two products arrive in beta, according to the company announcement from Cambridge, Massachusetts. Together they cover the two jobs that trip up teams moving AI out of demos and into production: serving a model reliably, and letting an agent run code without letting that code run wild.

Running a model in production takes more than picking a good one. A team needs serving infrastructure, access to its own data, routing, and rules about who can call what. Akamai Inference Agent Sandbox is built for that gap, with the inference side handling the serving and the sandbox side handling the execution. The company says enterprises can build and tune their own or hosted open-weight models, ground them in their own data, and secure the applications, agents, and infrastructure involved, all on one cloud platform.

Akamai Inference Agent Sandbox does not lock teams into one isolation design. The company describes agent isolation as converging on several models, from microVMs with snapshot and fork capability to policy-enforced containers to WebAssembly-isolated functions, and says developers can choose the architecture that fits their workload, including managed Kubernetes and NVIDIA-accelerated or dedicated processor infrastructure. The company also plans to add capabilities that govern what agents can reach and what can reach them, which matters because an agent that can browse, call tools, and touch internal systems is only as safe as the permissions around it.

What the inference service covers

Akamai Inference serves open-weight and custom models through managed application programming interfaces that follow the OpenAI format, so existing applications can move over without rewrites, according to the release. Customers can choose serverless pay-per-use inference or dedicated deployments. The launch catalog holds a dozen open-weight models spanning a wide size range, from compact models to very large ones, including entries from NVIDIA, OpenAI, and Google. Adam Karon, chief operating officer and general manager of Akamai Cloud, said in a statement that inference will become the main focus for most enterprises, and that customers want to build, run, and protect AI on the same platform.

Why agent code needs a cage

Agent Sandbox tackles the riskier half of the problem. Agents now write and execute their own code, and that code needs a hard boundary around what it can reach. The sandbox runs that code on Akamai Functions in its own isolated environment, separate from the agent runtime, using WebAssembly isolation to set limits on compute, memory, outbound connections, and system capabilities. Jon Alexander, senior vice president of cloud technology at Akamai, said in the release that every team running agents needs a boundary around the code those agents write.

The launch sits on infrastructure Akamai has been assembling for its AI push. The platform runs dense graphics and central processor clusters in core regions and reaches more than 4,350 edge locations, protected by the same security portfolio that defends many large online businesses. The company says that foundation has already secured multiyear contracts for Akamai Cloud collectively worth more than $14 billion. The release does not state pricing for the new products, and both remain in beta, so the open question is how quickly teams adopt the combined serving and sandbox approach over specialized AI infrastructure providers.

The timing fits a wider shift in how companies use AI. The pitch for Akamai Inference Agent Sandbox is that serving and sandboxing belong together, on infrastructure that already sits close to users and their data, instead of assembled from separate providers with the seams secured by hand. Whether teams buy that argument will show up in beta adoption, and in whether rival infrastructure providers answer with their own bundled isolation. For now, the products give Akamai a fuller AI stack to sell against the specialists, and give teams running agents a concrete way to answer the security questions that come with letting software write and run its own code.

For teams weighing Akamai Inference Agent Sandbox, the beta label is worth noting alongside the pitch. Akamai Inference Agent Sandbox is available now in beta, with serverless deployment options included in the inference beta, and the sandbox offered as an open source component of the Spin framework. Teams can test both without committing to a full migration, which is often how infrastructure bets like this get decided in practice.

Sources: Akamai announcement via GlobeNewswire and coverage of the beta launch.