diff --git a/src/__tests__/unit/checks/keywords-urls.test.ts b/src/__tests__/unit/checks/keywords-urls.test.ts
index 8f5cfa5..91ed9c5 100644
--- a/src/__tests__/unit/checks/keywords-urls.test.ts
+++ b/src/__tests__/unit/checks/keywords-urls.test.ts
@@ -32,6 +32,232 @@ describe('keywords guardrail', () => {
     expect(result.tripwireTriggered).toBe(false);
     expect(result.info?.matchedKeywords).toEqual([]);
   });
+
+  it('does not match partial words', () => {
+    const result = keywordsCheck(
+      {},
+      'Hello, world!',
+      KeywordsConfig.parse({ keywords: ['orld'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(false);
+  });
+
+  it('matches numbers', () => {
+    const result = keywordsCheck(
+      {},
+      'Hello, world123',
+      KeywordsConfig.parse({ keywords: ['world123'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(true);
+    expect(result.info?.matchedKeywords).toEqual(['world123']);
+  });
+
+  it('does not match partial numbers', () => {
+    const result = keywordsCheck(
+      {},
+      'Hello, world12345',
+      KeywordsConfig.parse({ keywords: ['world123'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(false);
+  });
+
+  it('matches underscores', () => {
+    const result = keywordsCheck(
+      {},
+      'Hello, w_o_r_l_d',
+      KeywordsConfig.parse({ keywords: ['w_o_r_l_d'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(true);
+    expect(result.info?.matchedKeywords).toEqual(['w_o_r_l_d']);
+  });
+
+  it('does not match when underscores appear inside other words', () => {
+    const result = keywordsCheck(
+      {},
+      'Hello, test_world_test',
+      KeywordsConfig.parse({ keywords: ['world'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(false);
+  });
+
+  it('matches chinese characters', () => {
+    const result = keywordsCheck(
+      {},
+      '你好',
+      KeywordsConfig.parse({ keywords: ['你好'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(true);
+  });
+
+  it('matches chinese characters with numbers', () => {
+    const result = keywordsCheck(
+      {},
+      '你好123',
+      KeywordsConfig.parse({ keywords: ['你好123'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(true);
+    expect(result.info?.matchedKeywords).toEqual(['你好123']);
+  });
+
+  it('does not match partial chinese characters with numbers', () => {
+    const result = keywordsCheck(
+      {},
+      '你好12345',
+      KeywordsConfig.parse({ keywords: ['你好123'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(false);
+  });
+
+  it('applies word boundaries across multi-keyword patterns', () => {
+    const result = keywordsCheck(
+      {},
+      'testing hello world',
+      KeywordsConfig.parse({ keywords: ['test', 'hello', 'world'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(true);
+    expect(result.info?.matchedKeywords).toEqual(['hello', 'world']);
+  });
+
+  it('matches keywords that start with special characters embedded in text', () => {
+    const result = keywordsCheck(
+      {},
+      'Reach me via example@foo.com later',
+      KeywordsConfig.parse({ keywords: ['@foo'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(true);
+    expect(result.info?.matchedKeywords).toEqual(['@foo']);
+  });
+
+  it('matches keywords that start with # even when preceded by letters', () => {
+    const result = keywordsCheck(
+      {},
+      'Use example#foo for the ID',
+      KeywordsConfig.parse({ keywords: ['#foo'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(true);
+    expect(result.info?.matchedKeywords).toEqual(['#foo']);
+  });
+
+  it('ignores keywords that become empty after sanitization', () => {
+    const result = keywordsCheck(
+      {},
+      'Totally benign text',
+      KeywordsConfig.parse({ keywords: ['!!!'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(false);
+    expect(result.info?.matchedKeywords).toEqual([]);
+    expect(result.info?.sanitizedKeywords).toEqual(['']);
+  });
+
+  it('still matches other keywords when some sanitize to empty strings', () => {
+    const result = keywordsCheck(
+      {},
+      'Please keep this secret!',
+      KeywordsConfig.parse({ keywords: ['...', 'secret!!!'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(true);
+    expect(result.info?.matchedKeywords).toEqual(['secret']);
+  });
+
+  it('matches keywords ending with special characters', () => {
+    const result = keywordsCheck(
+      {},
+      'Use foo@ in the config',
+      KeywordsConfig.parse({ keywords: ['foo@'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(true);
+    expect(result.info?.matchedKeywords).toEqual(['foo@']);
+  });
+
+  it('matches keywords ending with punctuation when followed by word characters', () => {
+    const result = keywordsCheck(
+      {},
+      'Check foo@example',
+      KeywordsConfig.parse({ keywords: ['foo@'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(true);
+    expect(result.info?.matchedKeywords).toEqual(['foo@']);
+  });
+
+  it('matches mixed script keywords', () => {
+    const result = keywordsCheck(
+      {},
+      'Welcome to hello你好world section',
+      KeywordsConfig.parse({ keywords: ['hello你好world'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(true);
+    expect(result.info?.matchedKeywords).toEqual(['hello你好world']);
+  });
+
+  it('does not match partial mixed script keywords', () => {
+    const result = keywordsCheck(
+      {},
+      'This is hello你好worldextra',
+      KeywordsConfig.parse({ keywords: ['hello你好world'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(false);
+  });
+
+  it('matches Arabic characters', () => {
+    const result = keywordsCheck(
+      {},
+      'مرحبا بك',
+      KeywordsConfig.parse({ keywords: ['مرحبا'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(true);
+    expect(result.info?.matchedKeywords).toEqual(['مرحبا']);
+  });
+
+  it('matches Cyrillic characters', () => {
+    const result = keywordsCheck(
+      {},
+      'Привет мир',
+      KeywordsConfig.parse({ keywords: ['Привет'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(true);
+    expect(result.info?.matchedKeywords).toEqual(['Привет']);
+  });
+
+  it('matches keywords with only punctuation', () => {
+    const result = keywordsCheck(
+      {},
+      'Use the @@ symbol',
+      KeywordsConfig.parse({ keywords: ['@@'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(true);
+    expect(result.info?.matchedKeywords).toEqual(['@@']);
+  });
+
+  it('matches mixed punctuation and alphanumeric keywords', () => {
+    const result = keywordsCheck(
+      {},
+      'Contact via @user123@',
+      KeywordsConfig.parse({ keywords: ['@user123@'] })
+    ) as GuardrailResult;
+
+    expect(result.tripwireTriggered).toBe(true);
+    expect(result.info?.matchedKeywords).toEqual(['@user123@']);
+  });
 });
 
 describe('urls guardrail', () => {
diff --git a/src/checks/keywords.ts b/src/checks/keywords.ts
index fe23ea6..6b37abf 100644
--- a/src/checks/keywords.ts
+++ b/src/checks/keywords.ts
@@ -40,6 +40,15 @@ export type KeywordsContext = z.infer<typeof KeywordsContext>;
  * @param config Configuration specifying keywords and behavior
  * @returns GuardrailResult indicating if tripwire was triggered
  */
+const WORD_CHAR_CLASS = '[\\p{L}\\p{N}_]';
+const isWordChar = (() => {
+  const wordCharRegex = new RegExp(WORD_CHAR_CLASS, 'u');
+  return (char: string | undefined): boolean => {
+    if (!char) return false;
+    return wordCharRegex.test(char);
+  };
+})();
+
 export const keywordsCheck: CheckFn<KeywordsContext, string, KeywordsConfig> = (
   ctx,
   text,
@@ -52,13 +61,40 @@ export const keywordsCheck: CheckFn<KeywordsContext, string, KeywordsConfig> = (
   // Sanitize keywords by stripping trailing punctuation
   const sanitizedKeywords = keywords.map((k: string) => k.replace(/[.,!?;:]+$/, ''));
 
-  // Create regex pattern with word boundaries
-  // Escape special regex characters and join with word boundaries
-  const escapedKeywords = sanitizedKeywords.map((k: string) =>
-    k.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')
-  );
-  const patternText = `\\b(?:${escapedKeywords.join('|')})\\b`;
-  const pattern = new RegExp(patternText, 'gi'); // case-insensitive, global
+  const keywordEntries = sanitizedKeywords
+    .map((sanitized) => ({
+      sanitized,
+      escaped: sanitized.replace(/[.*+?^${}()|[\]\\]/g, '\\$&'),
+    }))
+    .filter(({ sanitized }) => sanitized.length > 0);
+
+  if (keywordEntries.length === 0) {
+    return {
+      tripwireTriggered: false,
+      info: {
+        matchedKeywords: [],
+        originalKeywords: keywords,
+        sanitizedKeywords,
+        totalKeywords: keywords.length,
+        textLength: text.length,
+      },
+    };
+  }
+
+  // Apply unicode-aware word boundaries per keyword so tokens that start/end with punctuation still match.
+  const keywordPatterns = keywordEntries.map(({ sanitized, escaped }) => {
+    const keywordChars = Array.from(sanitized);
+    const firstChar = keywordChars[0];
+    const lastChar = keywordChars[keywordChars.length - 1];
+    const needsLeftBoundary = isWordChar(firstChar);
+    const needsRightBoundary = isWordChar(lastChar);
+    const leftBoundary = needsLeftBoundary ? `(?<!${WORD_CHAR_CLASS})` : '';
+    const rightBoundary = needsRightBoundary ? `(?!${WORD_CHAR_CLASS})` : '';
+    return `${leftBoundary}${escaped}${rightBoundary}`;
+  });
+
+  const patternText = `(?:${keywordPatterns.join('|')})`;
+  const pattern = new RegExp(patternText, 'giu'); // case-insensitive, global, unicode aware
 
   const matches: string[] = [];
   let match;