Anthropic lets companies run Claude Code on their own servers — a signal for regulated industries weighing AI coding tools
Anthropic has opened a public beta letting Team and Enterprise customers run Claude Code sessions on their own infrastructure instead of Anthropic's servers, addressing the data-residency and compliance objections that have kept regulated sectors like healthcare and finance cautious about AI coding agents.
17 August 2026
Anthropic has moved self-hosted environments for Claude Code into public beta, giving organisations on its Team and Enterprise plans the option to run coding-agent sessions on infrastructure they control rather than on Anthropic’s cloud. Sessions started from the web, desktop, terminal or a scheduled routine now run inside the organisation’s own network, next to its internal databases, registries and security controls, without those systems needing to be exposed to the public internet. Repository checkouts, build artefacts, secrets and anything a session creates stay on machines the organisation provisions. The feature is off by default and requires an admin to enable it; Anthropic recommends a platform or developer-experience team own the runner infrastructure, with a choice between a fixed pool of runners or an on-demand mode that scales with queued work.
This is a direct response to the objection that’s kept AI coding agents out of the most sensitive parts of regulated organisations: giving a third-party-hosted agent read/write access to a codebase touching patient data, financial records or government systems has been a hard sell for compliance teams, however useful the tool. Self-hosting doesn’t remove every concern — some early testers note prompts still round-trip to Anthropic’s models unless further isolated — but it materially changes the conversation from “can we use this at all” to “how do we configure it.”
So what
For teams in healthcare, finance or the public sector who’ve held off on AI-assisted development specifically over data-residency or compliance grounds, this is worth revisiting the decision over. It doesn’t remove the need for proper scoping of what any agent can access — self-hosted or not — but it removes one of the more defensible reasons to say no outright. If you’re weighing whether AI-assisted delivery is workable inside your compliance constraints, see our healthcare software development work or our AI-assisted development approach, or get in touch to talk through what’s actually feasible for your environment.