diff --git a/app/controllers/discourse_ai/admin/ai_spam_controller.rb b/app/controllers/discourse_ai/admin/ai_spam_controller.rb
index 4d4a39175..aae13cd10 100644
--- a/app/controllers/discourse_ai/admin/ai_spam_controller.rb
+++ b/app/controllers/discourse_ai/admin/ai_spam_controller.rb
@@ -11,11 +11,20 @@ def show
 
       def update
         updated_params = {}
-
-        updated_params[:llm_model_id] = allowed_params[:llm_model_id] if allowed_params.key?(
-          :llm_model_id,
-        )
-
+        if allowed_params.key?(:llm_model_id)
+          llm_model_id = updated_params[:llm_model_id] = allowed_params[:llm_model_id]
+          if llm_model_id.to_i < 0 &&
+               !SiteSetting.ai_spam_detection_model_allowed_seeded_models_map.include?(
+                 "custom:#{llm_model_id}",
+               )
+            return(
+              render_json_error(
+                I18n.t("discourse_ai.llm.configuration.invalid_seeded_model"),
+                status: 422,
+              )
+            )
+          end
+        end
         updated_params[:data] = {
           custom_instructions: allowed_params[:custom_instructions],
         } if allowed_params.key?(:custom_instructions)
diff --git a/app/serializers/ai_spam_serializer.rb b/app/serializers/ai_spam_serializer.rb
index 69edfd4a5..179d828ce 100644
--- a/app/serializers/ai_spam_serializer.rb
+++ b/app/serializers/ai_spam_serializer.rb
@@ -16,9 +16,9 @@ def custom_instructions
   end
 
   def available_llms
-    DiscourseAi::Configuration::LlmEnumerator.values.map do |hash|
-      { id: hash[:value], name: hash[:name] }
-    end
+    DiscourseAi::Configuration::LlmEnumerator
+      .values(allowed_seeded_llms: SiteSetting.ai_spam_detection_model_allowed_seeded_models_map)
+      .map { |hash| { id: hash[:value], name: hash[:name] } }
   end
 
   def flagging_username
diff --git a/lib/configuration/llm_enumerator.rb b/lib/configuration/llm_enumerator.rb
index 76f5b913a..9cbae76bb 100644
--- a/lib/configuration/llm_enumerator.rb
+++ b/lib/configuration/llm_enumerator.rb
@@ -50,14 +50,19 @@ def self.valid_value?(val)
         true
       end
 
-      def self.values
+      def self.values(allowed_seeded_llms: nil)
         values = DB.query_hash(<<~SQL).map(&:symbolize_keys)
           SELECT display_name AS name, id AS value
           FROM llm_models
         SQL
 
-        values.each { |value_h| value_h[:value] = "custom:#{value_h[:value]}" }
+        if allowed_seeded_llms.is_a?(Array)
+          values = values.filter do |value_h|
+            value_h[:value] > 0 || allowed_seeded_llms.include?("custom:#{value_h[:value]}")
+          end
+        end
 
+        values.each { |value_h| value_h[:value] = "custom:#{value_h[:value]}" }
         values
       end
 
diff --git a/spec/requests/admin/ai_spam_controller_spec.rb b/spec/requests/admin/ai_spam_controller_spec.rb
index df8e9172b..5f82d306d 100644
--- a/spec/requests/admin/ai_spam_controller_spec.rb
+++ b/spec/requests/admin/ai_spam_controller_spec.rb
@@ -25,6 +25,30 @@
         expect(AiModerationSetting.spam.data["custom_instructions"]).to eq("custom instructions")
       end
 
+      it "denies update for disallowed seeded llm" do
+        seeded_llm = Fabricate(:llm_model, id: -1)
+
+        put "/admin/plugins/discourse-ai/ai-spam.json",
+            params: {
+              is_enabled: true,
+              llm_model_id: seeded_llm.id,
+              custom_instructions: "custom instructions",
+            }
+
+        expect(response.status).to eq(422)
+
+        SiteSetting.ai_spam_detection_model_allowed_seeded_models = seeded_llm.identifier
+
+        put "/admin/plugins/discourse-ai/ai-spam.json",
+            params: {
+              is_enabled: true,
+              llm_model_id: seeded_llm.id,
+              custom_instructions: "custom instructions",
+            }
+
+        expect(response.status).to eq(200)
+      end
+
       it "can not enable spam detection without a model selected" do
         put "/admin/plugins/discourse-ai/ai-spam.json",
             params: {
@@ -163,13 +187,32 @@
     context "when logged in as admin" do
       before { sign_in(admin) }
 
-      it "returns the serialized spam settings" do
+      it "correctly filters seeded llms" do
         SiteSetting.ai_spam_detection_enabled = true
+        seeded_llm = Fabricate(:llm_model, id: -1, name: "seeded")
 
         get "/admin/plugins/discourse-ai/ai-spam.json"
+        expect(response.status).to eq(200)
+        json = response.parsed_body
 
+        # only includes fabricated model
+        expect(json["available_llms"].length).to eq(1)
+
+        SiteSetting.ai_spam_detection_model_allowed_seeded_models = seeded_llm.identifier
+
+        get "/admin/plugins/discourse-ai/ai-spam.json"
         expect(response.status).to eq(200)
+        json = response.parsed_body
 
+        expect(json["available_llms"].length).to eq(2)
+      end
+
+      it "returns the serialized spam settings" do
+        SiteSetting.ai_spam_detection_enabled = true
+
+        get "/admin/plugins/discourse-ai/ai-spam.json"
+
+        expect(response.status).to eq(200)
         json = response.parsed_body
         expect(json["is_enabled"]).to eq(true)
         expect(json["selected_llm"]).to eq(nil)