NEW REPORT | 451 RESEARCH BY S&P GLOBAL

As AI agents gain autonomy, who stays in control?

AI agents are transforming software development, but they’re also introducing new challenges around security, governance, and vendor dependence. In a new report, 451 Research examines how the Eclipse Foundation is addressing these challenges through open source innovation, vendor-neutral governance, and initiatives such as Eclipse Enclave and the Sovereign AI Foundation.

First page of the 451 Research report: Eclipse Foundation raises the governance and security stakes for AI agents

Two new bets on open AI governance

One brings organisations together to figure out where their AI dependencies are. The other gives developers a sandbox for the agents they already use.

Launched September 2026

A vendor-neutral consortium of 17 founding organisations from tech, manufacturing, aerospace and defense, and academia. Its job: help organisations keep practical control over their data, models, infrastructure and oversight.

  • Maps AI use cases and dependencies
  • Evaluates open source alternatives
  • Publishes white papers, blueprints and landscape analyses
  • Feeds enterprise requirements into Eclipse project roadmaps

Explore the Sovereign AI Foundation

Open source since May 2026, MIT licensed

A container-based sandbox for AI coding agents, built by EclipseSource as internal tooling and contributed to the Eclipse Foundation. Because the policy layer sits outside the agent, you can switch agents without rebuilding your setup.

  • Runs on Linux, macOS and WSL2
  • Works with Docker and rootless Podman
  • Extensible: share agents and policies with your team over Git

Claude Code, Codex CLI, Copilot CLI, Mistral, OpenCode, Eclipse Theia

Explore Eclipse Enclave

What Enclave locks down

The report looks at the four places an agent can cause trouble, and how Enclave controls each one.

network allowlisted state scoped commands controlled files isolated your-ai-agent
Eclipse Enclave puts a policy layer around the agent, whichever agent you run.
files

Which files the agent can read and write.

commands

Which commands it’s allowed to run.

network

Where it can connect, enforced by a DNS allowlist and proxy.

state

What carries over from one session to the next.

Credentials stay outside the agent’s container, and every allowed or blocked outbound request is logged for audit.

Enclave offers more immediate enterprise value.

The reason: one isolation and policy layer that works the same way across every supported AI coding agent, so switching agents doesn’t mean starting over.

Carl LehmannAnalyst, 451 Research by S&P Global

Built on open infrastructure that already runs at scale

450+open source projects
300+member organisations
~1Bmonthly Open VSX downloads, and climbing
17Sovereign AI Foundation founding organisations

What’s in the report

Why AI dependency is now a business risk

How agent autonomy, the EU AI Act and the Cyber Resilience Act are raising the governance bar.

How the pieces fit

Enclave and the Sovereign AI Foundation alongside PanEval, LMOS, Open VSX and the planned AI Registry.

The competitive picture

Where Enclave stands against vendor-built controls and DIY scripts, and how the Sovereign AI Foundation compares to other AI governance groups.

An honest SWOT

Including what still has to be proven: adoption, production references and steady delivery.