Info
Proposal is open for comment until 9 October
We particularly want to hear from teams already running agents against internal APIs, and from Security on the credential model.
ProposedAI
Agent execution sandbox
An isolated environment where agents can run code and call internal tools without holding standing credentials to production systems.
Lifecycle
Next: in development on 16 November 2026, in 6 weeks. Each audience arrives on its own schedule, below.
- Proposed (current stage)4 weeks ago
- In developmentplanned, in 6 weeks
- Rolling outplanned, in 5 months
- Generally availableplanned, in 8 months
By audience
Next up: Platform engineers building agents.
- Proposed
Platform engineers building agents
Phase 1 — agent builders
- Rolling out
- 1 Mar 2027
- Proposed
Teams already running agents against internal APIs
Phase 2 — existing internal-API agents
- Rolling out
- 10 May 2027
- Proposed
Any team, without asking us first
Phase 3 — self-serve
Detail
What is being proposed
A sandbox with no standing access: an agent requests a short-lived, scoped credential per task, and everything it runs is recorded.
Status
Proposed only. Scope and timing may change entirely, and it may not be built at all — the dates below are an estimate, not a commitment.