serialize-javascript escapes its output so it is safe to embed inside a
<script> element. In 7.1.1 that guarantee does not hold for function
values: a crafted function body can carry a literal, unescaped </script>
into the output, terminating the script element early so the remainder is
parsed as HTML.
SCRIPT_CLOSE_REGEXP used <\/script[^>]*> as its first alternative. The
character class excludes only >, so a single match could run from one
</script all the way to the next > anywhere in the source — swallowing a
second, complete </script> along the way. Only one replacement is emitted
per match, and the plain-code branch neutralizes just the leading <
('< ' + match.slice(1)), so the swallowed tag was re-emitted verbatim.
Reaching that shape requires </script in code position, which is legal
JavaScript: x</script=+/ parses as x < /script=+/, a comparison against a
regex literal.
const serialize = require('serialize-javascript');
const src = "function f(x){ return x</script=+/ + '</script><img src=x onerror=alert(1)>' }";
const out = serialize({ h: new Function('return ' + src)() });
// {"h":function f(x){ return x< /script=+/ + '</script><img src=x onerror=alert(1)>' }}
Embedded as the README documents (<script>window.S = <%= serialize(state) %></script>)
and parsed by Chromium, the script element ends at the injected tag and the
<img> becomes a live DOM node with its onerror handler executing in the
page origin.
Only the function path is affected. The same payload passed as data is
escaped correctly, and options.isJSON / non-function values are unaffected.
Fixed in 7.1.2. The wildcard now excludes < as well as >
([^<>]*), so a match can never reach past a second <. Every </script
in the source therefore either begins its own match or is followed by a
character the HTML tokenizer does not accept as ending a tag name — it ends
the tag name only on TAB, LF, FF, CR, SPACE, / or >, and emits anything
else as text.
Upgrade to 7.1.2. If you cannot upgrade, 7.1.0 and earlier are unaffected, or avoid serializing functions whose source text is attacker-influenced.
This is a regression specific to 7.1.1, not a long-standing issue. 7.1.0 and
earlier applied the same wildcard but escaped the entire match, so no tag
survived. Downstream scanners defaulting to a >= 7.1.0 range would be
overly broad.
{
"cwe_ids": [
"CWE-79",
"CWE-80"
],
"github_reviewed": true,
"github_reviewed_at": "2026-09-30T15:40:05Z",
"nvd_published_at": "2026-09-29T16:17:19Z",
"severity": "LOW"
}