diff --git a/HISTORY.md b/HISTORY.md
index b16fe57..959d7f7 100644
--- a/HISTORY.md
+++ b/HISTORY.md
@@ -4,6 +4,33 @@
 
 ### Bug Fixes
 
+* Sanitize now always removes `<noscript>` elements and their contents, even
+  when `noscript` is in the allowlist.
+
+  This fixes a sanitization bypass that could occur when `noscript` was allowed
+  by a custom allowlist. In this scenario, carefully crafted input could sneak
+  arbitrary HTML through Sanitize, potentially enabling an XSS (cross-site
+  scripting) attack.
+
+  Sanitize's default configs don't allow `<noscript>` elements and are not
+  vulnerable. This issue only affects users who are using a custom config that
+  adds `noscript` to the element allowlist.
+
+  The root cause of this issue is that HTML parsing rules treat the contents of
+  a `<noscript>` element differently depending on whether scripting is enabled
+  in the user agent. Nokogiri doesn't support scripting so it follows the
+  "scripting disabled" rules, but a web browser with scripting enabled will
+  follow the "scripting enabled" rules. This means that Sanitize can't reliably
+  make the contents of a `<noscript>` element safe for scripting enabled
+  browsers, so the safest thing to do is to remove the element and its contents
+  entirely.
+
+  See the following security advisory for additional details:
+  [GHSA-fw3g-2h3j-qmm7](https://github.com/rgrove/sanitize/security/advisories/GHSA-fw3g-2h3j-qmm7)
+
+  Thanks to David Klein from [TU Braunschweig](https://www.tu-braunschweig.de/en/ias)
+  (@leeN) for reporting this issue.
+
 * Fixed an edge case in which the contents of an "unescaped text" element (such
   as `<noembed>` or `<xmp>`) were not properly escaped if that element was
   allowlisted and was also inside an allowlisted `<math>` or `<svg>` element.
diff --git a/README.md b/README.md
index 0f6efb5..5f5e738 100644
--- a/README.md
+++ b/README.md
@@ -12,10 +12,10 @@ properties, @ rules, and URL protocols in elements or attributes containing CSS.
 Any HTML or CSS that you don't explicitly allow will be removed.
 
 Sanitize is based on the [Nokogumbo HTML5 parser][nokogumbo], which parses HTML
-exactly the same way modern browsers do, and [Crass][crass], which parses CSS
-exactly the same way modern browsers do. As long as your allowlist config only
-allows safe markup and CSS, even the most malformed or malicious input will be
-transformed into safe output.
+the same way modern browsers do, and [Crass][crass], which parses CSS the same
+way modern browsers do. As long as your allowlist config only allows safe markup
+and CSS, even the most malformed or malicious input will be transformed into
+safe output.
 
 [![Gem Version](https://badge.fury.io/rb/sanitize.svg)](http://badge.fury.io/rb/sanitize)
 [![Tests](https://github.com/rgrove/sanitize/workflows/Tests/badge.svg)](https://github.com/rgrove/sanitize/actions?query=workflow%3ATests)
@@ -427,6 +427,12 @@ elements not in this array will be removed.
 >
 > By default, Sanitize will remove all MathML and SVG elements. If you add MathML or SVG elements to a custom element allowlist, you must assume that any content inside them will be allowed, even if that content would otherwise be removed or escaped by Sanitize. This may create a security vulnerability in your application.
 
+> **Note**
+>
+> Sanitize always removes `<noscript>` elements and their contents, even if `noscript` is in the allowlist.
+>
+> This is because a `<noscript>` element's content is parsed differently in browsers depending on whether or not scripting is enabled. Since Nokogiri doesn't support scripting, it always parses `<noscript>` elements as if scripting is disabled. This results in edge cases where it's not possible to reliably sanitize the contents of a `<noscript>` element because Nokogiri can't fully replicate the parsing behavior of a scripting-enabled browser.
+
 #### :parser_options (Hash)
 
 [Parsing options](https://github.com/rubys/nokogumbo/tree/master#parsing-options) to be supplied to `nokogumbo`.
diff --git a/lib/sanitize/transformers/clean_element.rb b/lib/sanitize/transformers/clean_element.rb
index a2bacd8..98208a7 100644
--- a/lib/sanitize/transformers/clean_element.rb
+++ b/lib/sanitize/transformers/clean_element.rb
@@ -252,6 +252,16 @@ def call(env)
 
         node['content'] = node['content'].gsub(/;\s*charset\s*=.+\z/, ';charset=utf-8')
       end
+
+    # A `<noscript>` element's content is parsed differently in browsers
+    # depending on whether or not scripting is enabled. Since Nokogiri doesn't
+    # support scripting, it always parses `<noscript>` elements as if scripting
+    # is disabled. This results in edge cases where it's not possible to
+    # reliably sanitize the contents of a `<noscript>` element because Nokogiri
+    # can't fully replicate the parsing behavior of a scripting-enabled browser.
+    # The safest thing to do is to simply remove all `<noscript>` elements.
+    when 'noscript'
+      node.unlink
     end
   end
 
diff --git a/test/test_clean_element.rb b/test/test_clean_element.rb
index 1293cea..916b0f9 100644
--- a/test/test_clean_element.rb
+++ b/test/test_clean_element.rb
@@ -541,5 +541,12 @@
       )).must_equal "<html><head><meta http-equiv=\"Content-Type\" content=\"text/html;charset=utf-8\"></head><body>Howdy!</body></html>"
     end
 
+    it 'always removes `<noscript>` elements even if `noscript` is in the allowlist' do
+      assert_equal(
+        '',
+        Sanitize.fragment('<noscript>foo</noscript>', elements: ['noscript'])
+      )
+    end
+
   end
 end
diff --git a/test/test_malicious_html.rb b/test/test_malicious_html.rb
index 377e6de..7b33f68 100644
--- a/test/test_malicious_html.rb
+++ b/test/test_malicious_html.rb
@@ -244,7 +244,6 @@
     unescaped_content_elements = %w[
       noembed
       noframes
-      noscript
       plaintext
       script
       xmp
@@ -255,6 +254,7 @@
     ]
 
     removed_elements = %w[
+      noscript
       style
     ]
 
@@ -318,4 +318,22 @@
       end
     end
   end
+
+  describe 'sanitization bypass by exploiting scripting-disabled <noscript> behavior' do
+    before do
+      @s = Sanitize.new(
+        Sanitize::Config.merge(
+          Sanitize::Config::RELAXED,
+          elements: Sanitize::Config::RELAXED[:elements] + ['noscript']
+        )
+      )
+    end
+
+    it 'is prevented by removing `<noscript>` elements regardless of the allowlist' do
+      assert_equal(
+        '',
+        @s.fragment(%[<noscript><div id='</noscript>&lt;img src=x onerror=alert(1)&gt; '>])
+      )
+    end
+  end
 end