Mermaid CVE-2026-50159

Exploitation requires two simultaneous conditions: first, the application must render user-supplied Mermaid diagram source that includes attacker-controlled CSS properties - specifically via themeCSS, fontFamily, altFontFamily, or themeVariables config fields in the diagram definition; second, the rendered Mermaid <svg> element must have one or more sibling DOM elements in its parent container (i.e., it is not the only child). Default Mermaid integration via mermaid.run() or mermaid.initialize() places the SVG as an only child, which directly prevents exploitation - the vendor explicitly states 'most users of mermaid would not be affected.' Exploitation is limited to deployments where developers manually insert the SVG output into the DOM in a shared container. JavaScript execution is not achievable via this vector regardless of configuration. Authentication requirements are not confirmed from available data as no CVSS vector is provided.

No CVSS vector was provided with this CVE, so all metric assessments are inferred. The vulnerability is not listed in CISA KEV and no public exploit tooling exists, though the advisory publishes a functional proof-of-concept CSS payload. EPSS data is not available. Real-world risk is meaningfully constrained by the exploitation precondition: the SVG must have sibling DOM elements, which is explicitly a non-default condition - standard use of mermaid.run() or mermaid.initialize() places the SVG as an only child, blocking exploitation. This significantly narrows the affected population to applications that manually insert Mermaid SVG output into the DOM alongside other content. The impact ceiling is also bounded: no JavaScript execution, no session hijacking, no data exfiltration beyond conditional CSS-based probing. Risk is relevant to documentation platforms, wikis, or collaboration tools rendering untrusted Mermaid input, but is substantially lower than the 'RCE' tag suggests. The tag labeling in the source intelligence appears to be erroneous and should not drive prioritization.

An attacker submits a Mermaid diagram with a malicious themeCSS value - for example, & + * { position:fixed !important; inset:0 !important; background:red !important; width:100vw !important; height:100vh !important; } - to a platform that renders user-provided diagrams and inserts the SVG adjacent to page content. The sibling combinator in the injected CSS expands to target the next DOM sibling of the SVG, overlaying the entire page viewport with attacker-controlled styling, enabling phishing-style visual deception or suppression of legitimate page content. A functional proof-of-concept payload demonstrating this exact attack is published in the GitHub security advisory.