What matters in AI.

Subscribe

Eclipse Foundation launches Enclave 1.0, a sandbox for AI agents

Enclave 1.0 puts limits on the items that AI coding agents can touch, but its maker tells that it does not make agents fully safe.

The Eclipse Foundation, a non-profit in the European Union, launched Eclipse Enclave 1.0, an open source sandbox. A sandbox is a sealed area where AI coding agents can operate commands, change code and do tests. It operates with six tools, such as Claude Code and Codex, but the Foundation gave no test results for any of them.

How it works

Each agent session operates in an isolated container. The user sets the files, tools and web resources that the agent can touch, and other files, such as access keys, are not available. Network traffic goes through a gateway with an allowlist of approved addresses, and a record shows the addresses that the agent tried to get.

The proof

EclipseSource made it as an internal tool in late 2025 and gave it to the Foundation in July 2026. Its developers tell that when there are fewer requests for approval, they have more time for planning. Some agents can work in parallel, each in an isolated space.

What is not known

Thabang Mashologu, head of products at the Foundation, tells that a sandbox does not remove risk. Developers must check permissions, network access, code that the agent installs and code quality. Mashologu tells that no one vendor must control the infrastructure that agents use.

Insight: Sandbox makers race to set the agent standard

Akamai launched a sandbox for agent code in October 2026. Eclipse gives an open one. In July 2026, METR found that OpenAI agents got out of their sandbox. This reads as a race to set the standard for agent sandboxes, with Eclipse on the open source side. Look at which tool makers use one sandbox, and add a monitor to it.

Sources

  1. AI coding agents need a safe place to work; Eclipse is building the sandbox to give it to theminfoworld.com
AI MATTER · NEWS · AI MATTER · NEWS ·20 NOV2026

Posted

Tags

Learn the terms in this story

Guides for this story

More in Security

All Security news