Inclusion of Functionality from Untrusted Control Sphere vulnerability in the HTML5 scrubber in rrrene html_sanitize_ex allows a remote attacker to load a document of their choosing into a trusted page via the data attribute of an <object> element in sanitized HTML. object is the one URI-bearing element in lib/html_sanitize_ex/scrubber/html5.ex never registered through allow_tag_with_uri_attributes/3, and its only guard is a prefix match on lowercase "javascript:", so mixed-case variants, data: URIs, protocol-relative URLs and same-origin paths all survive.
This is not unconditional cross-site scripting. A javascript: URL does not execute through <object data> in current browsers, data: documents load in an opaque origin, and host-origin script execution additionally requires the application to serve attacker-controlled content from a same-origin path.
This issue affects html_sanitize_ex: from 0.3.1 before 1.5.3.
Only HtmlSanitizeEx.html5/1, and custom scrubbers declared with use HtmlSanitizeEx, extend: :html5, allow the object element, and the sanitized output must be rendered to other users.
Script execution in the host origin additionally requires the application to serve attacker-controlled HTML from a same-origin path, such as an uploads directory served as text/html.
Sanitize with basic_html/1, markdown_html/1 or strip_tags/1, none of which allow object, or define a custom scrubber that omits it.
A Content-Security-Policy of object-src 'none' blocks an injected <object> from loading, and hosting user uploads on a separate origin removes the escalation path.