diff --git a/exo/stdlib/requirements.py b/exo/stdlib/requirements.py
index 92e61f9..c1821f6 100644
--- a/exo/stdlib/requirements.py
+++ b/exo/stdlib/requirements.py
@@ -12,7 +12,9 @@
 
 from __future__ import annotations
 
+import io
 import re
+import tokenize
 from dataclasses import asdict, dataclass
 from pathlib import Path
 from typing import Any
@@ -321,6 +323,31 @@ def _scan_test_files(repo: Path, globs: list[str] | None = None) -> list[Path]:
     return sorted(found)
 
 
+def _scan_acc_in_python(file_path: Path, source: str) -> list[AccTestRef]:
+    """Scan a Python source for @acc: annotations using the tokenizer.
+
+    Only COMMENT tokens are checked so occurrences inside string literals
+    (e.g. test fixtures that quote the marker as data) are ignored.
+    TokenizeError is swallowed — the caller gets an empty list for that file.
+    """
+    refs: list[AccTestRef] = []
+    try:
+        tokens = list(tokenize.generate_tokens(io.StringIO(source).readline))
+    except tokenize.TokenError:
+        return refs
+    for tok in tokens:
+        if tok.type != tokenize.COMMENT:
+            continue
+        m = ACC_TAG_PATTERN.search(tok.string)
+        if m:
+            raw_ids = m.group(1)
+            for aid in raw_ids.split(","):
+                aid = aid.strip()
+                if aid:
+                    refs.append(AccTestRef(acc_id=aid, file=str(file_path), line=tok.start[0]))
+    return refs
+
+
 def scan_acc_refs(repo: Path, *, globs: list[str] | None = None) -> list[AccTestRef]:
     """Scan test files for @acc: annotations."""
     repo = Path(repo).resolve()
@@ -329,26 +356,24 @@ def scan_acc_refs(repo: Path, *, globs: list[str] | None = None) -> list[AccTest
 
     for filepath in files:
         try:
-            lines = filepath.read_text(encoding="utf-8", errors="replace").splitlines()
+            source = filepath.read_text(encoding="utf-8", errors="replace")
         except OSError:
             continue
 
         rel = str(filepath.relative_to(repo))
 
-        for line_num, line in enumerate(lines, start=1):
-            match = ACC_TAG_PATTERN.search(line)
-            if match:
-                raw_ids = match.group(1)
-                for aid in raw_ids.split(","):
-                    aid = aid.strip()
-                    if aid:
-                        refs.append(
-                            AccTestRef(
-                                acc_id=aid,
-                                file=rel,
-                                line=line_num,
-                            )
-                        )
+        if filepath.suffix == ".py":
+            for ref in _scan_acc_in_python(filepath, source):
+                refs.append(AccTestRef(acc_id=ref.acc_id, file=rel, line=ref.line))
+        else:
+            for line_num, line in enumerate(source.splitlines(), start=1):
+                match = ACC_TAG_PATTERN.search(line)
+                if match:
+                    raw_ids = match.group(1)
+                    for aid in raw_ids.split(","):
+                        aid = aid.strip()
+                        if aid:
+                            refs.append(AccTestRef(acc_id=aid, file=rel, line=line_num))
 
     return refs
 
diff --git a/tests/test_requirements.py b/tests/test_requirements.py
index 8a1f540..7119466 100644
--- a/tests/test_requirements.py
+++ b/tests/test_requirements.py
@@ -23,6 +23,7 @@
     REQ_TAG_PATTERN,
     VALID_PRIORITIES,
     VALID_STATUSES,
+    _scan_acc_in_python,
     format_req_trace_human,
     load_requirements,
     req_trace_to_dict,
@@ -1409,3 +1410,48 @@ def test_without_check_tests_flag(self, tmp_path: Path) -> None:
         data = json.loads(result.stdout)
         assert data["data"]["passed"]
         assert data["data"]["acc_total"] == 0
+
+
+# ── ACC Scanner String-Literal Immunity ─────────────────────────────
+
+
+class TestAccScannerStringLiteralImmunity:
+    """Tokenize-based scanner must not match @acc: inside Python string literals."""
+
+    def test_string_literal_yields_no_refs(self, tmp_path: Path) -> None:
+        """A @acc: marker embedded in a string literal must not be flagged."""
+        src = 'f = "# @acc: ACC-001"\n'
+        refs = _scan_acc_in_python(tmp_path / "test_fake.py", src)
+        assert refs == []
+
+    def test_real_comment_yields_one_ref(self, tmp_path: Path) -> None:
+        """A genuine @acc: comment must produce exactly one ref."""
+        src = "# @acc: ACC-002\ndef test_something(): pass\n"
+        refs = _scan_acc_in_python(tmp_path / "test_real.py", src)
+        assert len(refs) == 1
+        assert refs[0].acc_id == "ACC-002"
+
+    def test_mixed_source_yields_only_comment_ref(self, tmp_path: Path) -> None:
+        """Only the comment annotation counts; the string-literal one must be ignored."""
+        src = 'f = "# @acc: ACC-001"\n# @acc: ACC-002\ndef test_something(): pass\n'
+        refs = _scan_acc_in_python(tmp_path / "test_mixed.py", src)
+        assert len(refs) == 1
+        assert refs[0].acc_id == "ACC-002"
+
+    def test_tokenize_error_returns_empty(self, tmp_path: Path) -> None:
+        """Syntactically broken Python must not raise; scanner returns empty list."""
+        src = '"""\n'  # unterminated triple-quote forces tokenize.TokenError
+        refs = _scan_acc_in_python(tmp_path / "test_broken.py", src)
+        assert refs == []
+
+    def test_non_python_file_line_regex_preserved(self, tmp_path: Path) -> None:
+        """Non-Python files (.ts) must still be scanned via the line-regex path."""
+        repo = _bootstrap_repo(tmp_path)
+        (repo / "tests").mkdir(parents=True, exist_ok=True)
+        (repo / "tests" / "test_real.ts").write_text(
+            "// @acc: ACC-003\ntest('login', () => {});\n",
+            encoding="utf-8",
+        )
+        refs = scan_acc_refs(repo)
+        assert len(refs) == 1
+        assert refs[0].acc_id == "ACC-003"