Anthropic’s Claude Code just got a major upgrade for enterprise teams: Artifacts. On Claude Team and Enterprise subscription plans, a Claude Code session can now turn its work into a live, interactive, shareable, custom HTML webpage hosted at an interactive URL teammates can open and watch update in real time.

Here’s the immediate consequence. As Claude Code runs autonomously or under a user’s guidance, the connected webpage refreshes in-place at the same URL, updating charts and text instantly as code and connected data sources change. Anthropic also adds version history to each published artifact, so teammates can roll back or track progress securely on desktop or mobile. In other words, what used to be a transient terminal moment and a sprint-to-slack status update becomes something closer to an evolving internal product page.

This is not just a feature. It’s a “dynamic translation layer” between deep engineering work and the people who need to understand it. Anthropic frames Artifacts as built from the unbroken context of a user’s session: the agent uses the local repository codebase plus connected monitoring tools and conversational reasoning to spin up specialized web pages. Engineers do not need to wire up external data sources or stand up temporary infrastructure. The UI gets generated from what already exists, then stays linked to it. That matters because most teams waste time on glue work, not on the actual analysis, diagrams, or explanations.

The key detail is that these webpages are deliberately not full applications. Anthropic’s own documentation is blunt: “An artifact is a capture of work, not an application”. Each artifact is a single self-contained HTML page capped at a rendered size of 16 MiB. To keep enterprise security tight, Claude wraps the published file in a strict Content Security Policy (CSP) that blocks all external network requests. The practical effect is severe and intentional: the page cannot load external scripts, fonts, or stylesheets, and fetch, XHR, and WebSocket calls are blocked. All CSS and JavaScript must be inlined, and images must be embedded as data URIs.

That constraint will feel annoying to some builders, until you look at the compliance problem it solves. With artifacts, the output cannot call out to external services at view time, cannot store form input, cannot serve multiple routes, and cannot make API calls when someone views the page. In enterprise environments where the fear is “the AI will leak something” or “it will quietly reach the internet,” those limitations translate into a calmer security review. Anthropic is basically saying: we will help you publish dashboards and diagrams that update, but we will not turn your internal workflow into an externally connected web app.

That philosophical positioning becomes clearer when you compare it to OpenAI’s Codex Sites, which launched more than two weeks earlier. Both are aimed at enterprise hosting of AI work, but their shapes are different. OpenAI’s documentation describes Sites as platform-as-a-service for durable, full-stack web applications. Codex Sites output is compatible with Cloudflare Worker ES modules, and Sites supports persistent backend infrastructure. The documentation specifically calls out “D1” relational databases for structured data and “R2” object storage for file uploads. It also supports public sign-ins, identity provider integrations, and access controls tailored to workspace groups. Publishing is two-stage: save a reviewable candidate linked to a Git commit before official deployment. Codex Sites reads like a production environment meant to replace functional internal SaaS tools.

Claude Code Artifacts are the opposite, by design. They avoid backend capabilities, and Anthropic’s docs emphasize that artifacts are not applications. The security model matches that: every artifact is private to its author by default and cannot be made public to the broader internet. When an engineer shares a link, it is viewable exclusively by authenticated members of their specific organization. System administrators keep ultimate authority through org-level toggles, role-based scoping, and explicit retention policies, with oversight via a centralized compliance API.

OpenAI similarly emphasizes admin-gated access for Codex Sites, rolling the feature out primarily for ChatGPT Business and Enterprise workspaces. OpenAI requires system administrators to enable Sites via RBAC for Enterprise tiers. But because Sites behaves more like a hosted web application, access controls are more granular. When sharing a deployed URL, users can limit access to themselves and workspace admins, open it to all active users in the workspace, or restrict it to custom user groups. OpenAI also provides a dedicated Sites panel to manage runtime environment variables and secrets securely, so keys do not have to be committed to local source files.

If you zoom out, the business stakes look like this: AI is moving the “where work lives” question from chat threads to living surfaces people can monitor. That changes how teams collaborate, how progress gets audited, and how technical updates get consumed. It also affects how quickly non-technical stakeholders can get answers without asking engineers to rebuild dashboards manually.

And Anthropic is already pointing to actual use patterns. Boris Cherny, the Lead and creator of Claude Code, said in a post on X earlier today: “I’ve been using Artifacts in Claude Code for everything: visual explanations of tricky code, system diagrams, quick previews of a few animation options, data analyses and dashboards I share with the team,” adding that “They are a game changer for how I work with Claude.” The update also demonstrates the launch-material direction: turning “status updates” and “whiteboard diagrams” into secure, self-updating visual tools that do not leak live data outside the corporate boundary.

For enterprise decision-makers, the strategic question is no longer whether AI can generate code or reports. It’s whether your organization needs persistent application hosting like Codex Sites or secure, session-captured artifacts that behave like a controlled, internal publishing layer. Anthropic’s Artifacts press the advantage in speed, containment, and auditability of a specific workflow moment. OpenAI’s Sites press the advantage of durable infrastructure and full application replacement. The teams that win will be the ones that decide which layer they actually need, then align security, governance, and stakeholder expectations accordingly.