diff --git a/litellm/litellm_core_utils/llm_cost_calc/guardrail_cost.py b/litellm/litellm_core_utils/llm_cost_calc/guardrail_cost.py
index 64e82053c94..54cdf2cb8ff 100644
--- a/litellm/litellm_core_utils/llm_cost_calc/guardrail_cost.py
+++ b/litellm/litellm_core_utils/llm_cost_calc/guardrail_cost.py
@@ -36,16 +36,17 @@ class GuardrailCostByUnitEntry(BaseModel):
 
     model_config = ConfigDict(extra="ignore", frozen=True)
 
-    guardrail_cost_by_unit: Mapping[str, Annotated[float, Field(ge=0, allow_inf_nan=False)]] | None = None
+    guardrail_cost_by_unit: Mapping[str, Annotated[float, Field(ge=0, allow_inf_nan=False)] | None] | None = None
     guardrail_cost_in_spend: bool | None = True
 
 
 _GUARDRAIL_COST_BY_UNIT_ADAPTER: Final[TypeAdapter[GuardrailCostByUnitEntry]] = TypeAdapter(GuardrailCostByUnitEntry)
 
 
-def billed_guardrail_cost_by_unit(raw: object) -> Mapping[str, float] | None:
+def billed_guardrail_cost_by_unit(raw: object) -> Mapping[str, float | None] | None:
     """Per-counter USD the daily rollup may record for one raw ``guardrail_information``
-    entry; None when the entry is unpriced, report-only, or malformed."""
+    entry; None when the entry is unpriced, report-only, or malformed, and None per
+    counter the hook had no price for."""
     try:
         entry: Final = _GUARDRAIL_COST_BY_UNIT_ADAPTER.validate_python(raw)
     except ValidationError as e:
@@ -66,20 +67,28 @@ def _bedrock_guardrail_pricing(aws_region_name: str | None) -> GuardrailPricing
     return None
 
 
+def _priced_units(units: int, price_per_unit: float | None) -> float | None:
+    return None if price_per_unit is None else units * price_per_unit
+
+
 def bedrock_guardrail_cost_by_unit(
     usage_units: Mapping[str, int], aws_region_name: str | None
-) -> Mapping[str, float] | None:
-    """USD per counter, keyed like ``usage_units``; None when no pricing entry exists."""
+) -> Mapping[str, float | None] | None:
+    """USD per counter, keyed like ``usage_units``; None when no pricing entry exists,
+    and None for a counter the entry has no price for, since only an explicit 0.0 means free."""
     pricing: Final = _bedrock_guardrail_pricing(aws_region_name)
     if pricing is None:
         return None
     return {  # mutable-ok: stamped into guardrail_information, which safe_dumps only serializes as a plain dict
-        counter: units * pricing.guardrail_cost_per_unit.get(counter, 0.0) for counter, units in usage_units.items()
+        counter: _priced_units(units, pricing.guardrail_cost_per_unit.get(counter))
+        for counter, units in usage_units.items()
     }
 
 
-def guardrail_cost_total(cost_by_unit: Mapping[str, float] | None) -> float:
-    return sum(cost_by_unit.values()) if cost_by_unit is not None else 0.0
+def guardrail_cost_total(cost_by_unit: Mapping[str, float | None] | None) -> float:
+    """The scalar the spend path bills: unknown-priced counters count as 0 here, the
+    rollup keeps them unknown."""
+    return sum(cost for cost in cost_by_unit.values() if cost is not None) if cost_by_unit is not None else 0.0
 
 
 def bedrock_guardrail_cost(usage_units: Mapping[str, int], aws_region_name: str | None) -> float:
diff --git a/litellm/types/utils.py b/litellm/types/utils.py
index 8a6b1c13b2d..6c645226e2e 100644
--- a/litellm/types/utils.py
+++ b/litellm/types/utils.py
@@ -3142,10 +3142,11 @@ class StandardLoggingGuardrailInformation(TypedDict, total=False):
     provider hook. Summed into the request's ``response_cost`` so it counts against
     spend and budgets like token cost, unless ``guardrail_cost_in_spend`` is False."""
 
-    guardrail_cost_by_unit: ReadOnly[Mapping[str, float] | None]
+    guardrail_cost_by_unit: ReadOnly[Mapping[str, float | None] | None]
     """``guardrail_cost`` split per ``guardrail_usage`` counter, so the daily
     per-counter usage rollup can carry cost at its own grain. Absent when the
-    hook had no pricing for the invocation."""
+    hook had no pricing for the invocation; a counter is None when the pricing
+    entry has no price for it, which the rollup stores as unknown rather than $0."""
 
     guardrail_cost_in_spend: ReadOnly[bool | None]
     """Whether ``guardrail_cost`` participates in the request's ``response_cost`` and
@@ -3198,7 +3199,7 @@ class GuardrailTracingDetail(TypedDict, total=False):
     guardrail_action: str | None
     guardrail_usage: ReadOnly[Mapping[str, int] | None]
     guardrail_cost: ReadOnly[float | None]
-    guardrail_cost_by_unit: ReadOnly[Mapping[str, float] | None]
+    guardrail_cost_by_unit: ReadOnly[Mapping[str, float | None] | None]
     guardrail_cost_in_spend: ReadOnly[bool | None]
 
 
diff --git a/tests/test_litellm/litellm_core_utils/llm_cost_calc/test_guardrail_cost.py b/tests/test_litellm/litellm_core_utils/llm_cost_calc/test_guardrail_cost.py
index 6e9920d6f1d..af2f169157e 100644
--- a/tests/test_litellm/litellm_core_utils/llm_cost_calc/test_guardrail_cost.py
+++ b/tests/test_litellm/litellm_core_utils/llm_cost_calc/test_guardrail_cost.py
@@ -8,6 +8,7 @@ from litellm.litellm_core_utils.llm_cost_calc.guardrail_cost import (
     bedrock_guardrail_cost_by_unit,
     billed_guardrail_cost_by_unit,
     cost_breakdown_with_guardrail,
+    guardrail_cost_total,
     guardrail_information_cost,
 )
 
@@ -60,16 +61,19 @@ def test_bedrock_guardrail_cost_no_pricing_entry(monkeypatch):
 
 def test_bedrock_guardrail_cost_by_unit_prices_every_counter_it_was_given(synthetic_cost_map):
     """LIT-5652: the daily rollup stores one row per counter, so pricing must come
-    back at that grain, keyed exactly like the usage (free and unknown counters
-    included at 0.0) and summing to the scalar the spend path bills."""
+    back at that grain, keyed exactly like the usage. An explicit 0.0 in the cost
+    map is free; a counter the map does not list is unknown (None), never free,
+    while the scalar the spend path bills still sums only the known prices."""
     usage = {"contentPolicyUnits": 2, "topicPolicyUnits": 1, "wordPolicyUnits": 5, "someFutureCounter": 3}
     by_unit = bedrock_guardrail_cost_by_unit(usage_units=usage, aws_region_name="us-east-1")
     assert by_unit is not None
     assert by_unit.keys() == usage.keys()
     assert by_unit["contentPolicyUnits"] == pytest.approx(0.0003)
     assert by_unit["topicPolicyUnits"] == pytest.approx(0.00015)
-    assert (by_unit["wordPolicyUnits"], by_unit["someFutureCounter"]) == (0.0, 0.0)
-    assert sum(by_unit.values()) == pytest.approx(
+    assert by_unit["wordPolicyUnits"] == 0.0
+    assert by_unit["someFutureCounter"] is None
+    assert guardrail_cost_total(by_unit) == pytest.approx(0.00045)
+    assert guardrail_cost_total(by_unit) == pytest.approx(
         bedrock_guardrail_cost(usage_units=usage, aws_region_name="us-east-1")
     )
 
@@ -83,8 +87,15 @@ def test_bedrock_guardrail_cost_by_unit_is_none_without_pricing_so_unpriced_is_n
 
 
 def test_billed_guardrail_cost_by_unit_reads_the_hook_stamp():
-    entry = {"guardrail_name": "bedrock", "guardrail_cost_by_unit": {"contentPolicyUnits": 0.15, "wordPolicyUnits": 0}}
-    assert billed_guardrail_cost_by_unit(entry) == {"contentPolicyUnits": 0.15, "wordPolicyUnits": 0.0}
+    entry = {
+        "guardrail_name": "bedrock",
+        "guardrail_cost_by_unit": {"contentPolicyUnits": 0.15, "wordPolicyUnits": 0, "someFutureCounter": None},
+    }
+    assert billed_guardrail_cost_by_unit(entry) == {
+        "contentPolicyUnits": 0.15,
+        "wordPolicyUnits": 0.0,
+        "someFutureCounter": None,
+    }
 
 
 @pytest.mark.parametrize(
diff --git a/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_bedrock_guardrails.py b/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_bedrock_guardrails.py
index ec8996a8489..1ed24c9a59b 100644
--- a/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_bedrock_guardrails.py
+++ b/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_bedrock_guardrails.py
@@ -5095,18 +5095,31 @@ def test_build_tracing_detail_surfaces_usage_counters_and_cost(monkeypatch):
     detail = guardrail._build_tracing_detail(
         {
             "action": "GUARDRAIL_INTERVENED",
-            "usage": {"topicPolicyUnits": 1, "contentPolicyUnits": 2, "wordPolicyUnits": 0, "oddball": "not-an-int"},
+            "usage": {
+                "topicPolicyUnits": 1,
+                "contentPolicyUnits": 2,
+                "wordPolicyUnits": 0,
+                "someFutureCounter": 3,
+                "oddball": "not-an-int",
+            },
         },
         aws_region_name="us-east-1",
     )
 
-    assert detail["guardrail_usage"] == {"topicPolicyUnits": 1, "contentPolicyUnits": 2, "wordPolicyUnits": 0}
+    assert detail["guardrail_usage"] == {
+        "topicPolicyUnits": 1,
+        "contentPolicyUnits": 2,
+        "wordPolicyUnits": 0,
+        "someFutureCounter": 3,
+    }
     assert detail["guardrail_cost"] == pytest.approx(0.00045)
     by_unit = detail["guardrail_cost_by_unit"]
     assert by_unit is not None and by_unit.keys() == detail["guardrail_usage"].keys()
     assert by_unit["topicPolicyUnits"] == pytest.approx(0.00015)
     assert by_unit["contentPolicyUnits"] == pytest.approx(0.0003)
     assert by_unit["wordPolicyUnits"] == 0.0
+    assert by_unit["someFutureCounter"] is None
+    assert by_unit["wordPolicyUnits"] == 0.0
 
 
 def test_build_tracing_detail_omits_cost_by_unit_when_unpriced_but_keeps_scalar_zero(monkeypatch):
diff --git a/tests/test_litellm/proxy/guardrails/test_usage_tracking.py b/tests/test_litellm/proxy/guardrails/test_usage_tracking.py
index 50845385443..347c65cf819 100644
--- a/tests/test_litellm/proxy/guardrails/test_usage_tracking.py
+++ b/tests/test_litellm/proxy/guardrails/test_usage_tracking.py
@@ -373,6 +373,30 @@ async def test_cost_rolled_up_per_counter_alongside_units():
     assert costs["wordPolicyUnits"] == (0.0, {"increment": 0.0})
 
 
+@pytest.mark.asyncio
+async def test_counter_the_hook_could_not_price_is_stored_unknown_not_free():
+    """A counter the cost map does not list arrives stamped as None. Its row must
+    carry NULL, while the priced counter on the same request keeps its cost."""
+    prisma = _prisma()
+    logs = [
+        _payload(
+            "r1",
+            usage={"contentPolicyUnits": 1000, "someFutureCounter": 3},
+            cost_by_unit={"contentPolicyUnits": 0.15, "someFutureCounter": None},
+        )
+    ]
+
+    await process_spend_logs_guardrail_usage(prisma, logs)
+
+    assert _units_upserts(prisma) == {
+        ("bedrock-guard", "2026-08-17", "team-a", "hashed-key-1", "contentPolicyUnits"): 1000,
+        ("bedrock-guard", "2026-08-17", "team-a", "hashed-key-1", "someFutureCounter"): 3,
+    }
+    costs = _cost_upserts(prisma)
+    assert costs["contentPolicyUnits"] == (pytest.approx(0.15), {"increment": pytest.approx(0.15)})
+    assert costs["someFutureCounter"] == (None, None)
+
+
 @pytest.mark.asyncio
 async def test_unpriced_increment_makes_the_rows_cost_unknown_not_partial():
     """A payload with usage but no per-counter cost (a hook without pricing, a