ProseMirror View CVE-2026-104847

The application must embed prosemirror-view at a version prior to 1.42.3 as its editing surface, and a user must paste attacker-crafted HTML (containing slice-context attributes that are not run through schema attribute validation) into that editor. The decisive limiting factor is this required user interaction (CVSS UI:A): exploitation cannot occur without a victim performing a paste into the vulnerable editor, so it is not remotely triggerable without user action. Additional limits: only versions below 1.42.3 are affected, and the attacker must get their crafted content in front of a user likely to paste it into a ProseMirror-backed field.

This is a genuine, non-trivial client-side XSS (CWE-79) in a very widely-embedded editor library, but its real-world risk is moderated by a mandatory user-interaction gate. The vendor CVSS 4.0 vector (AV:N/AC:L/AT:N/PR:N/UI:A, VC:H/VI:H) correctly flags network reachability, low complexity, and no privileges, but UI:A is the decisive limiter: exploitation requires the victim to actively paste attacker-crafted HTML into a ProseMirror-based editor - it is not a drive-by or server-side flaw. No EPSS score, KEV entry, or public PoC was provided in the input, so there is no public exploit identified at time of analysis and no evidence of active exploitation; the only exploitation signal is the fix commit itself (credited to researcher Pedro Paniago). The high confidentiality/integrity impact is justified because successful exploitation runs attacker-controlled JavaScript in the page hosting the editor, enabling session theft or actions-as-user. Net: a legitimately high-severity library bug that is a real patch priority for anyone embedding prosemirror-view, but not an emergency mass-exploitation threat given the paste-based interaction requirement and absence of KEV/PoC/EPSS signals. No conflicting signals were detected in the provided data, though the absence of EPSS and PoC status means spread/weaponization cannot be assessed.

An attacker places specially crafted HTML on a page or in content a victim is likely to copy, then convinces the victim to paste it into a web application that embeds a ProseMirror editor prior to 1.42.3. During paste handling the clipboard slice context carries attributes that bypass schema attribute validation, and the resulting content executes attacker-controlled JavaScript in the victim's browser session. No public exploit was provided in the input, so this scenario is inferred from the description and the fix commit.