diff --git a/README.md b/README.md
index 6eec78b..8509291 100644
--- a/README.md
+++ b/README.md
@@ -123,6 +123,12 @@ tokens as well as optional ones.
 
 `TextFilterType.Fuzzy` does not take the setting: fuzzy matching is always case insensitive.
 
+Case is folded invariantly, so the answer does not depend on the calling thread's `CurrentCulture`.
+That matters for the dotted and dotless I: under a Turkish culture, `i` and `I` are not each other's
+case pair, and a culture-sensitive fold would stop a filter of `img` matching `IMG_1234.JPG` on a
+machine whose locale happens to be `tr-TR`. A filter pattern is machine text rather than prose, so
+it folds the same way everywhere.
+
 ### Filter Types
 
 TextFilter supports different filter types:
diff --git a/TextFilter.Test/TextFilter.Test.csproj b/TextFilter.Test/TextFilter.Test.csproj
index 135f735..87f0f7d 100644
--- a/TextFilter.Test/TextFilter.Test.csproj
+++ b/TextFilter.Test/TextFilter.Test.csproj
@@ -6,6 +6,14 @@
     <IsTestProject>true</IsTestProject>
     <TargetFramework>net10.0</TargetFramework>
     <TargetFrameworks></TargetFrameworks>
+    <!--
+      ktsu.Sdk defaults this to true, which resolves every culture to the invariant one and makes
+      new CultureInfo("tr-TR") throw. That default belongs to the application being built, not to
+      the applications that consume this library: a consumer outside the ktsu SDK runs with real
+      culture data, and culture-sensitive matching is only observable there. The test host needs
+      the culture data its callers have, or the tests that pin culture independence cannot run.
+    -->
+    <InvariantGlobalization>false</InvariantGlobalization>
   </PropertyGroup>
 
   <ItemGroup>
diff --git a/TextFilter.Test/TextFilterTests.cs b/TextFilter.Test/TextFilterTests.cs
index 310d096..397a5fc 100644
--- a/TextFilter.Test/TextFilterTests.cs
+++ b/TextFilter.Test/TextFilterTests.cs
@@ -3,6 +3,7 @@
 namespace TextFilter.Test;
 
 using System.Collections.Generic;
+using System.Globalization;
 using System.Linq;
 using ktsu.TextFilter;
 using Microsoft.VisualStudio.TestTools.UnitTesting;
@@ -607,6 +608,61 @@ public void TheTwoSensitivitiesDoNotCollideInTheGlobCache()
 		Assert.IsFalse(TextFilter.IsMatch("B.MD", "*.md", TextFilterType.Glob, TextFilterMatchOptions.ByWholeString, TextFilterCaseSensitivity.CaseSensitive));
 	}
 
+	[TestMethod]
+	public void RegexCaseInsensitivityDoesNotDependOnTheCurrentCulture()
+	{
+		// Turkish folds "i" to "İ" and "I" to "ı", so IgnoreCase without CultureInvariant stops
+		// treating "i" and "I" as the same letter. Only a pattern containing that letter shows it:
+		// ".*\.jpg" matches "IMG_1234.JPG" across case under tr-TR perfectly well, which is why the
+		// case-insensitivity tests above never caught this.
+		CultureInfo original = CultureInfo.CurrentCulture;
+		try
+		{
+			CultureInfo.CurrentCulture = new CultureInfo("tr-TR");
+
+			// The premise, stated rather than assumed: under globalization-invariant mode this
+			// culture request silently resolves to the invariant culture, and the assertion below
+			// would then pass without ever exercising the fold it is about.
+			if (CultureInfo.CurrentCulture.TextInfo.ToUpper("i") == "I")
+			{
+				Assert.Inconclusive("This runtime does not apply Turkish case mapping, so the regression cannot be provoked here.");
+			}
+
+			Assert.IsTrue(
+				TextFilter.IsMatch("IMG_5678.JPG", "img", TextFilterType.Regex, TextFilterMatchOptions.ByWholeString, TextFilterCaseSensitivity.CaseInsensitive),
+				"A case-insensitive filter should fold i and I whatever locale the calling thread runs under.");
+		}
+		finally
+		{
+			CultureInfo.CurrentCulture = original;
+		}
+	}
+
+	[TestMethod]
+	public void ARegexCompiledUnderOneCultureAnswersTheSameUnderAnother()
+	{
+		// The cache is keyed by pattern and sensitivity, not by culture, so a culture-dependent fold
+		// is not merely wrong under tr-TR -- it leaks. Whichever culture compiled the pattern first
+		// decides the answer for every later caller on every thread. A pattern used nowhere else, so
+		// this owns its cache entry and the ordering below is the one that actually runs.
+		CultureInfo original = CultureInfo.CurrentCulture;
+		try
+		{
+			CultureInfo.CurrentCulture = new CultureInfo("tr-TR");
+			bool underTurkish = TextFilter.IsMatch("MINI.TXT", "mini", TextFilterType.Regex, TextFilterMatchOptions.ByWholeString, TextFilterCaseSensitivity.CaseInsensitive);
+
+			CultureInfo.CurrentCulture = CultureInfo.InvariantCulture;
+			bool underInvariant = TextFilter.IsMatch("MINI.TXT", "mini", TextFilterType.Regex, TextFilterMatchOptions.ByWholeString, TextFilterCaseSensitivity.CaseInsensitive);
+
+			Assert.IsTrue(underTurkish, "The first caller's culture should not decide the answer.");
+			Assert.AreEqual(underTurkish, underInvariant, "A cached pattern should answer the same for every caller.");
+		}
+		finally
+		{
+			CultureInfo.CurrentCulture = original;
+		}
+	}
+
 	[TestMethod]
 	public void TheTwoSensitivitiesDoNotCollideInTheRegexCache()
 	{
diff --git a/TextFilter/TextFilter.cs b/TextFilter/TextFilter.cs
index 1cf1c68..df64d37 100644
--- a/TextFilter/TextFilter.cs
+++ b/TextFilter/TextFilter.cs
@@ -408,8 +408,14 @@ public static bool DoesMatchRegex(string text, string filter, TextFilterMatchOpt
 		string cacheKey = CacheKey(filter, caseSensitivity);
 		if (!RegexCache.TryGetValue(cacheKey, out Regex? regex))
 		{
+			// CultureInvariant, because IgnoreCase alone folds case using the thread's CurrentCulture.
+			// Under tr-TR that stops "i" and "I" being the same letter, so a filter of "img" no longer
+			// matches "IMG_1234.JPG" -- and the cache below is keyed by pattern and sensitivity only,
+			// so whichever culture happened to compile the pattern first decides the answer for every
+			// later caller on any thread. A filter pattern is machine text, not prose, and the glob
+			// path already folds invariantly, so the two paths now agree.
 			RegexOptions regexOptions = caseSensitivity is TextFilterCaseSensitivity.CaseInsensitive
-				? RegexOptions.Compiled | RegexOptions.IgnoreCase
+				? RegexOptions.Compiled | RegexOptions.IgnoreCase | RegexOptions.CultureInvariant
 				: RegexOptions.Compiled;
 
 			try