Proactively block free models when data collection is not allowed

apps/web/src/app/api/openrouter/[...path]/route.ts

@@ -28,9 +28,11 @@ import {
   isDeadFreeModel,
   isExcludedForFeature,
   isKiloExclusiveFreeModel,
-  isKiloExclusiveModelRequiringDataCollection,
 } from '@/lib/ai-gateway/models';
-import { isFreeModel } from '@/lib/ai-gateway/is-free-model';
+import {
+  hasBestEffortGuessDataCollectionRequirement,
+  isFreeModel,
+} from '@/lib/ai-gateway/is-free-model';
 import {
   accountForMicrodollarUsage,
   captureProxyError,
@@ -554,17 +556,8 @@ export async function POST(request: NextRequest): Promise<NextResponseType<unkno
   }
   let effectiveProviderContext = initialProviderResultForAbuseService;
 
-  // Request-level data-collection opt-out: a caller can set
-  // `provider.data_collection: 'deny'` or `provider.zdr: true` on any
-  // request to opt that single request out of training/data-retention.
-  // Direct experiment upstreams ignore those OpenRouter/Vercel flags
-  // (we never reach OpenRouter), but we still capture the prompt to R2
-  // for partner evaluation — which violates the caller's stated
-  // intent. Refuse here regardless of org settings, anon/BYOK status,
-  // or the org-level check below.
   if (
-    (effectiveProviderContext.experiment ||
-      isKiloExclusiveModelRequiringDataCollection(effectiveModelIdLowerCased)) &&
+    (await hasBestEffortGuessDataCollectionRequirement(effectiveModelIdLowerCased)) &&
     isDataCollectionExplicitlyDisallowed(requestBodyParsed.body.provider)
   ) {
     return dataCollectionRequiredResponse();

apps/web/src/lib/ai-gateway/is-free-model.ts

@@ -1,5 +1,9 @@
 import { KILO_AUTO_FREE_MODEL } from '@/lib/ai-gateway/auto-model';
-import { isKiloExclusiveFreeModel, isOpenRouterStealthModel } from '@/lib/ai-gateway/models';
+import {
+  isKiloExclusiveFreeModel,
+  isOpenRouterStealthModel,
+  kiloExclusiveModels,
+} from '@/lib/ai-gateway/models';
 import { isPublicIdExperimented } from '@/lib/ai-gateway/experiments/membership';
 
 /**
@@ -22,3 +26,15 @@ export async function isFreeModel(model: string): Promise<boolean> {
     (await isPublicIdExperimented(model ?? ''))
   );
 }
+
+export async function hasBestEffortGuessDataCollectionRequirement(model: string): Promise<boolean> {
+  return (
+    (await isFreeModel(model)) ||
+    kiloExclusiveModels.some(
+      candidate =>
+        candidate.public_id === model &&
+        candidate.status !== 'disabled' &&
+        candidate.flags.includes('requires-data-collection')
+    )
+  );
+}

apps/web/src/lib/ai-gateway/models.test.ts

@@ -1,11 +1,6 @@
 import { describe, test, expect } from '@jest/globals';
-import {
-  autoFreeModels,
-  findKiloExclusiveModel,
-  kiloExclusiveModels,
-  isKiloExclusiveModelRequiringDataCollection,
-} from './models';
-import { isFreeModel } from './is-free-model';
+import { autoFreeModels, findKiloExclusiveModel, kiloExclusiveModels } from './models';
+import { hasBestEffortGuessDataCollectionRequirement, isFreeModel } from './is-free-model';
 import { getInferenceProvider } from './providers/kilo-exclusive-model';
 import {
   claude_opus_4_7_stealth_model,
@@ -75,18 +70,6 @@ describe('isFreeModel', () => {
       expect(claude_opus_4_6_stealth_model.public_id).toBe('stealth/claude-opus-4.6');
     });
 
-    test('requires data collection for paid training-enabled offerings', () => {
-      expect(
-        isKiloExclusiveModelRequiringDataCollection(claude_opus_4_7_stealth_model.public_id)
-      ).toBe(true);
-      expect(
-        isKiloExclusiveModelRequiringDataCollection(claude_sonnet_4_6_stealth_model.public_id)
-      ).toBe(true);
-      expect(
-        isKiloExclusiveModelRequiringDataCollection(claude_opus_4_6_stealth_model.public_id)
-      ).toBe(true);
-    });
-
     test('all Kilo exclusive models should have either no pricing or valid ordered pricing tiers', () => {
       for (const model of kiloExclusiveModels) {
         if (model.pricing) {
@@ -186,3 +169,27 @@ describe('isFreeModel', () => {
     });
   });
 });
+
+describe('hasBestEffortGuessDataCollectionRequirement', () => {
+  test('requires data collection for paid training-enabled offerings', async () => {
+    expect(
+      await hasBestEffortGuessDataCollectionRequirement(claude_opus_4_7_stealth_model.public_id)
+    ).toBe(true);
+    expect(
+      await hasBestEffortGuessDataCollectionRequirement(claude_sonnet_4_6_stealth_model.public_id)
+    ).toBe(true);
+    expect(
+      await hasBestEffortGuessDataCollectionRequirement(claude_opus_4_6_stealth_model.public_id)
+    ).toBe(true);
+  });
+
+  test('requires data collection for free models', async () => {
+    expect(await hasBestEffortGuessDataCollectionRequirement('openrouter/free')).toBe(true);
+  });
+
+  test('does not require data collection for regular paid models', async () => {
+    expect(await hasBestEffortGuessDataCollectionRequirement('anthropic/claude-sonnet-4')).toBe(
+      false
+    );
+  });
+});

apps/web/src/lib/ai-gateway/models.ts

@@ -93,15 +93,6 @@ export const kiloExclusiveModels = [
   stepfun_37_flash_free_model,
 ] as KiloExclusiveModel[];
 
-export function isKiloExclusiveModelRequiringDataCollection(model: string): boolean {
-  return kiloExclusiveModels.some(
-    m =>
-      m.public_id === model &&
-      m.status !== 'disabled' &&
-      (!m.pricing || m.flags.includes('requires-data-collection'))
-  );
-}
-
 export function isKiloStealthModel(model: string): boolean {
   return kiloExclusiveModels.some(m => m.public_id === model && m.flags.includes('stealth'));
 }