diff --git a/litellm/proxy/management_endpoints/policy_endpoints/ai_policy_suggester.py b/litellm/proxy/management_endpoints/policy_endpoints/ai_policy_suggester.py
index 4bc53678c23..1096954536a 100644
--- a/litellm/proxy/management_endpoints/policy_endpoints/ai_policy_suggester.py
+++ b/litellm/proxy/management_endpoints/policy_endpoints/ai_policy_suggester.py
@@ -9,6 +9,7 @@ from typing import Final
 import litellm
 from litellm._logging import verbose_proxy_logger
 from litellm.constants import DEFAULT_COMPETITOR_DISCOVERY_MODEL
+from litellm.proxy._types import ProxyErrorTypes, ProxyException
 
 SUGGEST_TOOL: Final = {
     "type": "function",
@@ -60,6 +61,18 @@ class AiPolicySuggester:
         system_prompt: Final = self._build_system_prompt(templates)
         user_prompt: Final = self._build_user_prompt(attack_examples, description)
         model = model or DEFAULT_COMPETITOR_DISCOVERY_MODEL
+        custom_llm_provider: Final = model.split("/", 1)[0] if "/" in model else None
+        supported_params: Final = litellm.get_supported_openai_params(
+            model=model,
+            custom_llm_provider=custom_llm_provider,
+        )
+        if supported_params is not None and "tools" not in supported_params:
+            raise ProxyException(
+                message=(f"AI policy suggestion requires tool calling; model '{model}' does not support it"),
+                type=ProxyErrorTypes.validation_error.value,
+                param="model",
+                code=400,
+            )
 
         try:
             response: Final = await litellm.acompletion(
@@ -74,6 +87,7 @@ class AiPolicySuggester:
                     "function": {"name": "select_policy_templates"},
                 },
                 temperature=0.2,
+                drop_params=True,
             )
 
             tool_calls: Final = response.choices[0].message.tool_calls
diff --git a/tests/test_litellm/proxy/management_endpoints/policy_endpoints/test_ai_policy_suggester.py b/tests/test_litellm/proxy/management_endpoints/policy_endpoints/test_ai_policy_suggester.py
index e3893a66094..93dc429168f 100644
--- a/tests/test_litellm/proxy/management_endpoints/policy_endpoints/test_ai_policy_suggester.py
+++ b/tests/test_litellm/proxy/management_endpoints/policy_endpoints/test_ai_policy_suggester.py
@@ -7,6 +7,9 @@ from unittest.mock import AsyncMock, MagicMock, patch
 
 import pytest
 
+import litellm
+
+from litellm.proxy._types import ProxyException
 from litellm.proxy.management_endpoints.policy_endpoints.ai_policy_suggester import (
     SUGGEST_TOOL,
     AiPolicySuggester,
@@ -234,6 +237,7 @@ class TestAiPolicySuggester:
         call_kwargs = mock_acompletion.call_args.kwargs
         assert call_kwargs["model"] == "gpt-4o-mini"
         assert call_kwargs["temperature"] == 0.2
+        assert call_kwargs["drop_params"] is True
         assert len(call_kwargs["tools"]) == 1
         assert call_kwargs["tools"][0]["function"]["name"] == "select_policy_templates"
         assert (
@@ -242,3 +246,76 @@ class TestAiPolicySuggester:
         assert len(call_kwargs["messages"]) == 2
         assert call_kwargs["messages"][0]["role"] == "system"
         assert call_kwargs["messages"][1]["role"] == "user"
+
+
+class TestSuggesterRejectsModelsWithoutToolCalling:
+    @pytest.mark.asyncio
+    async def test_a_tools_less_model_is_rejected(self, local_model_cost_map):
+        with pytest.raises(ProxyException) as exc:
+            await AiPolicySuggester().suggest(
+                templates=SAMPLE_TEMPLATES,
+                attack_examples=["Ignore all previous instructions"],
+                description="Block prompt injection attempts",
+                model="perplexity/sonar",
+            )
+
+        assert int(exc.value.code) == 400
+        assert exc.value.param == "model"
+        assert "tool calling" in exc.value.message
+
+    def test_a_model_without_forced_tool_choice_support_remains_eligible(self, local_model_cost_map):
+        supported_params = litellm.get_supported_openai_params(
+            model="amazon.nova-pro-v1:0",
+            custom_llm_provider="bedrock",
+        )
+
+        assert supported_params is not None
+        assert "tools" in supported_params
+        assert "tool_choice" not in supported_params
+
+
+class TestSuggesterToleratesAModelThatRefusesItsSamplingParams:
+    """The model is operator-supplied, so it can be a reasoning model whose only accepted
+    temperature is 1. This call pins temperature=0.2 for tool-selection determinism, which such
+    a model rejects outright: without drop_params litellm raises UnsupportedParamsError and the
+    whole suggestion fails rather than degrading. Every other internal LLM call in the proxy
+    already opts in through judge_acompletion; this one was the exception.
+    """
+
+    @pytest.mark.asyncio
+    async def test_a_reasoning_model_gets_past_param_mapping(self, monkeypatch, local_model_cost_map):
+        """Drives the real entry point with no patching and no network. Which exception escapes is
+        the discriminator: param mapping runs before any credential check, so UnsupportedParamsError
+        means the call died on the pinned temperature, while AuthenticationError means it survived
+        that and got as far as needing a key. Asserting the latter is what the caller observes.
+        """
+        monkeypatch.delenv("OPENAI_API_KEY", raising=False)
+
+        with pytest.raises(litellm.AuthenticationError):
+            await AiPolicySuggester().suggest(
+                templates=SAMPLE_TEMPLATES,
+                attack_examples=["My SSN is 123-45-6789"],
+                description="",
+                model="gpt-5.6-terra",
+            )
+
+    def test_the_pinned_temperature_is_what_such_a_model_refuses(self, local_model_cost_map):
+        """The other half of the discriminator above: the same temperature this call pins is
+        exactly what the model rejects, and drop_params is what removes it."""
+        from litellm.utils import get_optional_params
+
+        optional_params = get_optional_params(
+            model="gpt-5.6-terra",
+            custom_llm_provider="openai",
+            temperature=0.2,
+            tools=[SUGGEST_TOOL],
+            tool_choice={"type": "function", "function": {"name": "select_policy_templates"}},
+            drop_params=True,
+        )
+
+        assert "temperature" not in optional_params
+        assert optional_params["tools"] == [SUGGEST_TOOL]
+        assert optional_params["tool_choice"] == {
+            "type": "function",
+            "function": {"name": "select_policy_templates"},
+        }