diff --git a/providers/common/ai/docs/changelog.rst b/providers/common/ai/docs/changelog.rst
index dd5daeb30472e..5fc707ae7c1d7 100644
--- a/providers/common/ai/docs/changelog.rst
+++ b/providers/common/ai/docs/changelog.rst
@@ -37,12 +37,14 @@ Changelog
   :doc:`retry_policies`, "When the connection also carries a fallback chain".
 
 .. note::
-  ``SQLToolset(allowed_tables=[])`` now raises ``ValueError``. Up to 0.9.0 an empty list
-  was accepted and exposed every table in the schema -- the same as ``allowed_tables=None``
-  -- so a Dag that builds the list dynamically (a ``Variable.get``, a config file, a
-  filtered comprehension) silently handed the agent the whole schema whenever the list
-  came back empty. Such a Dag now fails at import instead. Pass ``None`` explicitly if
-  exposing every table is what you meant.
+  ``SQLToolset`` now rejects ``allowed_tables=None`` and ``allowed_tables=[]`` with
+  ``ValueError``. Up to 0.9.0 both were accepted and exposed every table in the schema, so
+  a Dag that builds the list dynamically (a ``Variable.get`` with a ``None`` default, a
+  config file, a filtered comprehension) silently handed the agent the whole schema
+  whenever the lookup came back empty. Such a Dag now fails at import instead. Exposing
+  every table is still the default, but only by omitting the argument: no value you can
+  pass requests it, so a runtime lookup can never widen the allow-list by accident. Dags
+  that passed ``allowed_tables=None`` explicitly should drop the argument.
 
 0.9.0
 .....
diff --git a/providers/common/ai/docs/choosing_a_toolset.rst b/providers/common/ai/docs/choosing_a_toolset.rst
index 3ac32937e331b..cc7f8a54a2af7 100644
--- a/providers/common/ai/docs/choosing_a_toolset.rst
+++ b/providers/common/ai/docs/choosing_a_toolset.rst
@@ -144,11 +144,12 @@ subqueries and joins.
   something other than a DBAPI cursor — ``ExasolHook`` and its pyexasol
   statement, for instance — fall back to a full fetch. The payload handed to the
   model is still bounded; the transfer is not. See :ref:`bounded-query-results`.
-- Its parser-level closure is opt-in, not the default. ``allowed_tables``
-  defaults to ``None`` and the table walk returns immediately while it is
-  unset, so out of the box the agent reaches every table the connection can
-  see. ``DESCRIBE`` and ``SHOW`` both pass, on dialects that parse them,
-  while ``allowed_tables`` stays unset. Set ``allowed_tables`` and the walk
+- Its parser-level closure is opt-in, not the default. The table walk returns
+  immediately while ``allowed_tables`` is unset, so out of the box the agent
+  reaches every table the connection can see. Only omitting the argument grants
+  that: an explicit ``None`` or empty list is rejected at construction.
+  ``DESCRIBE`` and ``SHOW`` both pass, on dialects that parse them, while
+  ``allowed_tables`` stays unset. Set ``allowed_tables`` and the walk
   turns fail-closed for ``SHOW`` — but not for ``DESCRIBE``: it instead
   becomes an ordinary table reference, allowed only when the table it names
   is on the list. See :ref:`allowed-tables-enforcement` for what else the
diff --git a/providers/common/ai/docs/toolsets.rst b/providers/common/ai/docs/toolsets.rst
index 3461935fe62f7..63cbfa0139972 100644
--- a/providers/common/ai/docs/toolsets.rst
+++ b/providers/common/ai/docs/toolsets.rst
@@ -206,9 +206,11 @@ Parameters
 ^^^^^^^^^^
 
 - ``db_conn_id``: Airflow connection ID for the database.
-- ``allowed_tables``: Restrict the agent to a fixed set of tables. ``None``
-  (default) exposes all tables in ``schema``; an empty list raises ``ValueError``
-  rather than silently exposing them all. Entries may be schema-qualified
+- ``allowed_tables``: Restrict the agent to a fixed set of tables. Omit the
+  argument (the default) to expose all tables in ``schema``. No value means
+  allow-all: ``None`` and an empty list both raise ``ValueError``, so an allow-list
+  built at runtime that resolves to nothing fails at import instead of silently
+  exposing every table. Entries may be schema-qualified
   (``"SCHEMA.TABLE"``) to span multiple schemas; see above. Matching is
   case-insensitive. When set, the list is enforced on ``query`` and
   ``check_query`` as well as discovery -- every table a query references must be
diff --git a/providers/common/ai/src/airflow/providers/common/ai/toolsets/sql.py b/providers/common/ai/src/airflow/providers/common/ai/toolsets/sql.py
index 5011ba99c2095..bbfd32108b661 100644
--- a/providers/common/ai/src/airflow/providers/common/ai/toolsets/sql.py
+++ b/providers/common/ai/src/airflow/providers/common/ai/toolsets/sql.py
@@ -52,6 +52,13 @@
 if TYPE_CHECKING:
     from pydantic_ai._run_context import RunContext
 
+# Sentinel distinguishing "caller did not pass ``allowed_tables``" (expose every
+# table) from an explicit value. Every explicit falsy value -- ``None`` as much as
+# ``[]`` -- is rejected, so an allow-list a Dag builds at runtime can never widen to
+# allow-all by resolving to nothing. Allow-all is reachable only by omitting the
+# argument, which is a static, visible decision in the Dag file.
+_UNSET: Any = object()
+
 # JSON Schemas for the four SQL tools.
 _LIST_TABLES_SCHEMA: dict[str, Any] = {
     "type": "object",
@@ -170,9 +177,12 @@ class SQLToolset(AbstractToolset[Any]):
     toolset does not inspect the error type or message.
 
     :param db_conn_id: Airflow connection ID for the database.
-    :param allowed_tables: Restrict the agent to a fixed set of tables. ``None``
-        (default) exposes every table in ``schema``; an empty list raises
-        ``ValueError`` rather than silently exposing every table. Entries may be
+    :param allowed_tables: Restrict the agent to a fixed set of tables. Omit the
+        argument (the default) to expose every table in ``schema``. No *value* means
+        allow-all: ``None`` and an empty list both raise ``ValueError``, so an allow-list
+        built at runtime (a ``Variable.get`` with a ``None`` default, a config lookup, a
+        filtered comprehension) that resolves to nothing fails at import instead of
+        silently handing the agent the whole schema. Entries may be
         schema-qualified (``"SCHEMA.TABLE"``) to span multiple schemas in one database
         -- common on warehouses such as Snowflake. ``list_tables`` introspects each referenced
         schema and returns the matching tables fully qualified, and ``get_schema``
@@ -240,20 +250,26 @@ def __init__(
         self,
         db_conn_id: str,
         *,
-        allowed_tables: list[str] | None = None,
+        allowed_tables: list[str] = _UNSET,
         allowed_functions: list[str] | None = None,
         schema: str | None = None,
         allow_writes: bool = False,
         max_rows: int = 50,
         max_result_bytes: int = DEFAULT_MAX_RESULT_BYTES,
     ) -> None:
-        if allowed_tables is not None and not allowed_tables:
+        self._allowed_tables: frozenset[str] | None
+        if allowed_tables is _UNSET:
+            self._allowed_tables = None
+        elif not allowed_tables:
             raise ValueError(
-                "allowed_tables must not be empty. Pass None to allow every table in the schema, "
-                "or list the tables the agent may access."
+                f"allowed_tables must name at least one table, got {allowed_tables!r}. Omit the "
+                "argument to expose every table in the schema. An empty or missing value is "
+                "rejected rather than read as 'no restriction' so that an allow-list built at "
+                "runtime cannot silently widen to every table."
             )
+        else:
+            self._allowed_tables = frozenset(allowed_tables)
         self._db_conn_id = db_conn_id
-        self._allowed_tables: frozenset[str] | None = frozenset(allowed_tables) if allowed_tables else None
         # Case-folded so matching a query's function names (also case-folded) is
         # case-insensitive, mirroring how allowed_tables is compared.
         self._allowed_functions: frozenset[str] = (
diff --git a/providers/common/ai/tests/unit/common/ai/toolsets/test_sql.py b/providers/common/ai/tests/unit/common/ai/toolsets/test_sql.py
index 455805d104364..57e0db38b0d93 100644
--- a/providers/common/ai/tests/unit/common/ai/toolsets/test_sql.py
+++ b/providers/common/ai/tests/unit/common/ai/toolsets/test_sql.py
@@ -96,9 +96,15 @@ def test_id_includes_conn_id(self):
         ts = SQLToolset("my_pg")
         assert ts.id == "sql-my_pg"
 
-    def test_empty_allowed_tables_raises(self):
-        with pytest.raises(ValueError, match="allowed_tables must not be empty"):
-            SQLToolset("my_pg", allowed_tables=[])
+    @pytest.mark.parametrize("value", [None, []], ids=["none", "empty-list"])
+    def test_falsy_allowed_tables_raises(self, value):
+        """Only omitting the argument grants allow-all; no explicit value does."""
+        with pytest.raises(ValueError, match="allowed_tables must name at least one table"):
+            SQLToolset("my_pg", allowed_tables=value)
+
+    def test_omitted_allowed_tables_means_no_restriction(self):
+        ts = SQLToolset("my_pg")
+        assert ts._allowed_tables is None
 
 
 class TestSQLToolsetGetTools: