diff --git a/ddtrace/internal/utils/formats.py b/ddtrace/internal/utils/formats.py
index 1fd5f007e4a..02dd5381b06 100644
--- a/ddtrace/internal/utils/formats.py
+++ b/ddtrace/internal/utils/formats.py
@@ -63,7 +63,7 @@ def asbool(value: Union[str, bool, None]) -> bool:
     return value.lower() in ("true", "1")
 
 
-def parse_tags_str(tags_str: Optional[str]) -> dict[str, str]:
+def parse_tags_str(tags_str: Optional[str], sep: Optional[str] = None) -> dict[str, str]:
     """
     Parses a string containing key-value pairs and returns a dictionary.
     Key-value pairs are delimited by ':', and pairs are separated by whitespace, comma, OR BOTH.
@@ -71,13 +71,16 @@ def parse_tags_str(tags_str: Optional[str]) -> dict[str, str]:
     This implementation aligns with the way tags are parsed by the Agent and other Datadog SDKs
 
     :param tags_str: A string of the above form to parse tags from.
+    :param sep: An explicit pair separator to use instead of auto-detecting one. Callers whose
+        values may themselves contain whitespace (and no comma) should pass "," here to avoid an
+        incorrect whitespace-based split.
     :return: A dict containing the tags that were parsed.
     """
     res: dict[str, str] = {}
     if not tags_str:
         return res
     # falling back to comma as separator
-    sep = "," if "," in tags_str else " "
+    sep = sep if sep is not None else ("," if "," in tags_str else " ")
 
     for tag in tags_str.split(sep):
         tag = tag.strip()
diff --git a/ddtrace/llmobs/_writer.py b/ddtrace/llmobs/_writer.py
index 0783ca90e76..6909008915e 100644
--- a/ddtrace/llmobs/_writer.py
+++ b/ddtrace/llmobs/_writer.py
@@ -216,7 +216,10 @@ def __init__(
             self._headers[EVP_SUBDOMAIN_HEADER_NAME] = self.EVP_SUBDOMAIN_HEADER_VALUE
         additional_header_str = env.get("_DD_TRACE_WRITER_ADDITIONAL_HEADERS", "")
         if additional_header_str:
-            self._headers.update(parse_tags_str(additional_header_str))
+            # Explicit comma separator: header values (e.g. "Bearer <token>") may contain a space
+            # with no comma anywhere in the string, which would otherwise fall back to a whitespace
+            # split and corrupt the value.
+            self._headers.update(parse_tags_str(additional_header_str, sep=","))
 
         self._send_payload_with_retry = fibonacci_backoff_with_jitter(
             attempts=self.RETRY_ATTEMPTS,
diff --git a/releasenotes/notes/fix-additional-headers-whitespace-split-53067051dd9a93f2.yaml b/releasenotes/notes/fix-additional-headers-whitespace-split-53067051dd9a93f2.yaml
new file mode 100644
index 00000000000..409fef2b3d2
--- /dev/null
+++ b/releasenotes/notes/fix-additional-headers-whitespace-split-53067051dd9a93f2.yaml
@@ -0,0 +1,6 @@
+---
+fixes:
+  - |
+    LLM Observability: fixes an issue where ``_DD_TRACE_WRITER_ADDITIONAL_HEADERS`` header values
+    containing a whitespace (for example ``Bearer <token>``) were incorrectly parsed, resulting in
+    a truncated header value that caused LLM Observability spans to be dropped.
diff --git a/tests/llmobs/test_llmobs_span_agentless_writer.py b/tests/llmobs/test_llmobs_span_agentless_writer.py
index bcc06ef54e4..907a9964ebb 100644
--- a/tests/llmobs/test_llmobs_span_agentless_writer.py
+++ b/tests/llmobs/test_llmobs_span_agentless_writer.py
@@ -37,6 +37,38 @@ def test_additional_headers(mock_writer_logs):
     assert llmobs_span_writer._headers["Authorization"] == "Bearer-custom-token"
 
 
+@pytest.mark.parametrize(
+    "additional_headers,expected_headers",
+    [
+        # No comma: would wrongly fall back to a whitespace split.
+        (
+            "Authorization:Bearer custom-token",
+            {"Authorization": "Bearer custom-token"},
+        ),
+        # Trailing comma.
+        (
+            "Authorization:Bearer custom-token,",
+            {"Authorization": "Bearer custom-token"},
+        ),
+        # Multiple real-world headers, two with spaces in their value.
+        (
+            'Authorization:Bearer abc123xyz,User-Agent:"Datadog Tracer",X-Request-Id:req-12345',
+            {
+                "Authorization": "Bearer abc123xyz",
+                "User-Agent": '"Datadog Tracer"',
+                "X-Request-Id": "req-12345",
+            },
+        ),
+    ],
+)
+def test_additional_headers_with_spaces(mock_writer_logs, additional_headers, expected_headers):
+    assert expected_headers
+    with mock.patch.dict(os.environ, {"_DD_TRACE_WRITER_ADDITIONAL_HEADERS": additional_headers}):
+        llmobs_span_writer = LLMObsSpanWriter(1, 1, is_agentless=True, _site=DD_SITE, _api_key=DD_API_KEY)
+    for key, value in expected_headers.items():
+        assert llmobs_span_writer._headers[key] == value
+
+
 def test_no_additional_headers_by_default(mock_writer_logs):
     llmobs_span_writer = LLMObsSpanWriter(1, 1, is_agentless=True, _site=DD_SITE, _api_key=DD_API_KEY)
     assert "Authorization" not in llmobs_span_writer._headers
diff --git a/tests/tracer/test_utils.py b/tests/tracer/test_utils.py
index 06643c78a7c..486269dd290 100644
--- a/tests/tracer/test_utils.py
+++ b/tests/tracer/test_utils.py
@@ -67,6 +67,20 @@ def test_parse_env_tags(tag_str, expected_tags):
     assert parse_tags_str(tag_str) == expected_tags, tag_str
 
 
+@pytest.mark.parametrize(
+    "tag_str,sep,expected_tags",
+    [
+        # No comma anywhere: auto-detection would fall back to whitespace and split "Bearer <token>".
+        # An explicit "," separator keeps it as a single value.
+        ("Authorization:Bearer abc123xyz", ",", {"Authorization": "Bearer abc123xyz"}),
+        # Explicit " " separator forces a whitespace split even though a comma is present.
+        ("a:b,c bKey:bVal", " ", {"a": "b,c", "bKey": "bVal"}),
+    ],
+)
+def test_parse_env_tags_explicit_sep(tag_str, sep, expected_tags):
+    assert parse_tags_str(tag_str, sep=sep) == expected_tags, tag_str
+
+
 @pytest.mark.parametrize(
     "key,value,expected",
     [