Security and sanitization

Treat model output as untrusted input. HyperMarkdown defaults to html="sanitize": raw HTML is parsed and cleaned before math, highlighting, diagrams, or animation run.

Scripts, styles, iframes, forms, and event-handler attributes are removed.

HTML policies

Mode	Behavior
sanitize	Default. Parse HTML and remove anything outside the schema.
literal	Show raw markup as text. Strongest choice for untrusted output.
raw	Parse without sanitization. Only for content you control.

<HyperMarkdown md={markdown} html="literal" />

sanitize={false} remains for compatibility and selects raw mode when html is not provided. Prefer the explicit html prop.

Extend the schema

<HyperMarkdown
  md={markdown}
  allowedTags={{ mention: ["data-user-id"] }}
  linkSafety={{
    allowedLinkPrefixes: ["https://docs.example.com/"],
    allowedImagePrefixes: ["https://images.example.com/"],
    allowDataImages: false,
  }}
/>

By default, http, https, mailto, and tel protocols are permitted, along with data images. Narrow those defaults when your product has a stricter trust boundary.

Sanitization is not a substitute for controlling what external URLs a browser may request. Combine link safety with your Content Security Policy and image proxy where appropriate.