Claude's Artifacts, Styles, and File Search Move to VM Execution by June 15

Anthropic notified Team and Enterprise administrators in May that three long-standing Claude features will not survive the summer in their current form. Legacy Artifacts, custom writing Styles, and Project file upload search are all scheduled for retirement by June 15. Their successors exist and work differently, and each requires a feature that many enterprise deployments have not yet turned on: code execution in a sandboxed virtual machine environment. After the deadline, organizations that have not enabled code execution will lose access to all three capabilities.

Three Features, Three Replacements

The first change covers Artifacts, the outputs Claude generates as self-contained files, such as HTML pages, data tables, and code. The legacy Artifacts system ran directly in the browser environment. Its replacement runs in an isolated sandbox container, which Anthropic calls VM-based Artifacts. The new architecture supports a feature the old one could not: Live Artifacts, dashboards that pull from connected data sources and refresh in real time without requiring users to re-prompt Claude. Anthropic says existing artifacts migrate automatically; administrators only need to enable code execution before June 15 for the migration to succeed.

The second change affects Styles, the saved writing-personality profiles many teams have used to enforce consistent tone across Claude outputs. Under the new model, saved Styles convert to Skills, which are structured packages of instructions and scripts that execute inside the VM environment. Organizations that prefer not to use VM-based Skills can convert their Style definitions to plain Claude instructions instead, stored at the organization level in the Claude console. Both migration paths require an administrator to act before the cutover date.

The third change involves how Claude searches files uploaded to a Project. The current implementation uses vector search to retrieve relevant passages and surface them to the model. The replacement uses the VM's code execution layer to search and reason over project files directly, which Anthropic says improves accuracy on complex multi-document queries. Like the other two changes, it depends on code execution being enabled.

One Architecture, Across the Platform

The simultaneous migration of three separate features to the same underlying execution model signals something broader about Anthropic's platform direction. Over the past year, the company has assembled a single architecture for agentic work: Claude Managed Agents, sandboxed code execution, Skills, and now VM-based Artifacts and file search. The June 15 consolidation folds the remaining user-facing features into that architecture, replacing what were previously heterogeneous implementations with a common execution layer.

The practical consequence for enterprise IT is that code execution, a permission many organizations have treated as a narrow developer feature, is becoming load-bearing infrastructure for Claude's core productivity capabilities. Organizations that have disabled it on security grounds will need to revisit that policy before the deadline. Anthropic's documentation describes the sandbox containers as isolated from users' systems, with no network access outside explicitly granted connections. The relevant question for security teams is whether that isolation model meets their policies for dynamic code execution.

"The permission you have been deferring is no longer optional if you want Claude's core productivity features to keep working after June 15." Developer summary of Anthropic's Team and Enterprise admin notice, May 2026

What to Do Before the Deadline

The administrative checklist is short. Log in to the Claude Team or Enterprise console and verify that code execution is enabled for your organization. If it is not enabled, turn it on and test your most-used workflows in the new environment before June 15. For Styles, decide whether to migrate to Skills or convert to plain instructions, then complete that migration in the console; existing Style content does not need to be re-entered, but the conversion step is manual. For Projects with uploaded files, no action is needed beyond enabling code execution, since the file search replacement activates automatically.

For organizations running Claude in regulated environments with strict execution controls, the larger task is the security review. Approving a new class of sandboxed code execution typically requires sign-off from security teams that may need lead time to review Anthropic's documentation on the VM isolation model. Leaving that review to the last week of the migration window is risky for larger organizations.

The broader context is that Anthropic has been moving steadily toward a platform where all of Claude's dynamic capabilities run inside controlled sandbox environments rather than in ad-hoc browser or server contexts. The Claude Managed Agents and self-hosted sandboxes announcement from May follows the same logic. So does the Claude Platform on AWS launch, which brought Managed Agents webhooks and multiagent orchestration to AWS infrastructure this week. June 15 is less a one-time migration event and more a milestone in that broader consolidation.

Anthropic has confirmed that Team and Enterprise admins received direct notification of the change. Organizations that have not yet reviewed that notice should do so now. The June 15 deadline does not offer an extension path for teams that miss it, and restoring access to the affected features after the cutover will require enabling code execution regardless, so the timeline for compliance is fixed.