diff --git a/CHANGELOG.md b/CHANGELOG.md
index c18e8da84..fb2a8101a 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,7 @@
 # Changelog
 
+- Three proxy paths no longer relay upstream or JS error text to the client. The HUD summary passed OpenAI's own `error.message` through whenever upstream was not ok, carrying request ids and quota wording; the Realtime token route echoed the JS error, which for a network fault names the upstream host; and a failed CCTV media fetch stored the raw errno as the camera's health message, which reaches the screen through `GET /api/cctv/health` rather than through the sanitized response beside it. Logs now name the failure and the upstream status without the text.
+
 - Separate vessel records and feed acquisition from rendering while preserving selection, partial-feed retention, sea-surface placement and request cancellation.
 
 - Separate military-flight records and acquisition from rendering while preserving ground-model ownership, source units and follow behavior.
diff --git a/docs/CURRENT-STATE.md b/docs/CURRENT-STATE.md
index ee9ee6015..ee5458a07 100644
--- a/docs/CURRENT-STATE.md
+++ b/docs/CURRENT-STATE.md
@@ -3192,6 +3192,7 @@ are omitted rather than framing the wrong part of the globe.
 - A cold OpenSky failure uses the current camera subpoint only to request a cached adsb.lol point fallback capped at 250 nm. A fresh OpenSky response or last-good cache wins; a nominally successful worldwide snapshot more than two minutes old prefers viewport-scoped adsb.lol when available, otherwise the stale source is reported honestly. The fallback is visibly source-labeled and is never presented as a worldwide snapshot.
 - GBFS proxy refuses upstream redirects (`redirect: 'manual'`; any 3xx becomes a 502 and the redirect target is logged server-side only) and enforces its 5 MB response cap while the body streams, cancelling the upstream read past the cap; CCTV health map is bounded.
 - Proxy error payloads are sanitized (no internal error details returned to clients).
+- That holds for the OpenAI and CCTV media paths too: `/api/openai/hud-summary` never relays OpenAI's own `error.message`, `/api/realtime/token` answers a fixed string when minting fails, and a failed CCTV media fetch stores a fixed camera health `message` — `GET /api/cctv/health` serializes that field and the CCTV panel renders it as a status label, so it is a client surface as much as the response body is.
 - `OPENAI_API_KEY` is server-side only; the browser receives ephemeral Realtime client secrets from `/api/realtime/token`.
 - `AISSTREAM_API_KEY` is server-side only; the browser reads the same-origin `/api/ais-live` cache.
 - `/api/google/nearby-places` keeps the Google key out of Places requests issued for voice scene context.
diff --git a/server/providers/cctv.js b/server/providers/cctv.js
index 5f601532f..69512e7c5 100644
--- a/server/providers/cctv.js
+++ b/server/providers/cctv.js
@@ -308,11 +308,16 @@ export function cctvProxy({ sourceRoot = process.cwd() } = {}) {
             }
             const timedOut =
               error?.name === 'AbortError' || error?.name === 'TimeoutError';
+            // GET /api/cctv/health serializes `message`, and the CCTV panel
+            // renders it as a status label, so the raw error would leave the
+            // server by a different door than the sanitized body below and
+            // land on screen. Keep the detail in the log.
+            console.warn('[CCTV Proxy] media fetch failed');
             setHealth(cameraId, {
               status: 'degraded',
               sourceKind: 'upstream',
               label: source?.provider || 'Configured source',
-              message: error?.message || 'Media fetch failed',
+              message: 'Media fetch failed',
             });
             res.writeHead(timedOut ? 504 : 502, {
               'Content-Type': 'application/json',
diff --git a/server/providers/openai/hud-summary.js b/server/providers/openai/hud-summary.js
index 54c065412..92d38f2e8 100644
--- a/server/providers/openai/hud-summary.js
+++ b/server/providers/openai/hud-summary.js
@@ -78,20 +78,23 @@ async function handleHudSummary(req, res) {
     res.statusCode = response.ok && summary ? 200 : response.status || 502;
     res.setHeader('Content-Type', 'application/json; charset=utf-8');
     res.setHeader('Cache-Control', 'no-store');
+    if (!response.ok)
+      console.warn(`[hud-summary] upstream HTTP ${response.status}`);
     res.end(
       JSON.stringify({
         summary: summary || null,
-        error: response.ok
-          ? null
-          : data.error?.message || 'OpenAI HUD summary request failed',
+        // Never relay `data.error.message`: that is OpenAI's own wording, and
+        // it carries request ids, organization hints and quota phrasing.
+        error: response.ok ? null : 'OpenAI HUD summary request failed',
       }),
     );
-  } catch (error) {
+  } catch {
+    console.warn('[hud-summary] request failed');
     res.statusCode = 502;
     res.setHeader('Content-Type', 'application/json');
     res.end(
       JSON.stringify({
-        error: error?.message || 'OpenAI HUD summary request failed',
+        error: 'OpenAI HUD summary request failed',
       }),
     );
   }
diff --git a/server/providers/openai/realtime.js b/server/providers/openai/realtime.js
index 53c9d60da..74a2eccb9 100644
--- a/server/providers/openai/realtime.js
+++ b/server/providers/openai/realtime.js
@@ -146,12 +146,15 @@ function createRealtimeTokenHandler({
         res.setHeader('X-GEV-Voice-Tier-Fallback', '1');
       }
       res.end(body);
-    } catch (error) {
+    } catch {
+      // For a network fault this was a resolver message naming the upstream
+      // host; the client only needs to know the mint failed.
+      console.warn('[realtime-token] mint failed');
       res.statusCode = 502;
       res.setHeader('Content-Type', 'application/json');
       res.end(
         JSON.stringify({
-          error: error?.message || 'Failed to create Realtime token',
+          error: 'Failed to create Realtime token',
         }),
       );
     }
diff --git a/src/tooling/localServices.test.mjs b/src/tooling/localServices.test.mjs
index d7f2e4ba3..395dd4f09 100644
--- a/src/tooling/localServices.test.mjs
+++ b/src/tooling/localServices.test.mjs
@@ -283,3 +283,36 @@ test('Realtime service configuration selects compatible endpoint/model without f
   assert.equal(response.headers['cache-control'], 'no-store');
   assert.doesNotMatch(response.body, /server-fixture|voice\.example/);
 });
+
+test('OpenAI routes answer generically when the upstream or the request fails', async (t) => {
+  env(t, 'OPENAI_API_KEY', 'fixture-upstream-secret');
+  env(t, 'GEV_RATELIMIT_OPENAI_PER_MIN', undefined);
+  const leak =
+    'fixture-upstream-secret req_fixture_1234 org-fixture quota exhausted';
+
+  // `data.error.message` is OpenAI's own wording — request ids, organization
+  // hints, quota phrasing — and was relayed verbatim whenever upstream was not ok.
+  t.mock.method(globalThis, 'fetch', async () =>
+    Response.json({ error: { message: leak } }, { status: 429 }),
+  );
+  const summary = await request(
+    install(openAiRealtimeProxy()).get('/api/openai/hud-summary'),
+    { method: 'POST', body: JSON.stringify({ context: {} }) },
+  );
+  assert.equal(summary.json().error, 'OpenAI HUD summary request failed');
+  assert.equal(summary.body.includes('req_fixture_1234'), false);
+  assert.equal(summary.body.includes('fixture-upstream-secret'), false);
+
+  // A network fault surfaced a resolver message naming the upstream host.
+  t.mock.restoreAll();
+  t.mock.method(globalThis, 'fetch', async () => {
+    throw Error(`getaddrinfo ENOTFOUND api.openai.com ${leak}`);
+  });
+  const token = await request(
+    install(openAiRealtimeProxy()).get('/api/realtime/token'),
+  );
+  assert.equal(token.status, 502);
+  assert.deepEqual(token.json(), { error: 'Failed to create Realtime token' });
+  assert.equal(token.body.includes('api.openai.com'), false);
+  assert.equal(token.body.includes('fixture-upstream-secret'), false);
+});
diff --git a/src/tooling/mediaProviders.test.mjs b/src/tooling/mediaProviders.test.mjs
index cbfcd0624..9642f55b6 100644
--- a/src/tooling/mediaProviders.test.mjs
+++ b/src/tooling/mediaProviders.test.mjs
@@ -149,3 +149,45 @@ for (const hook of ['configureServer', 'configurePreviewServer']) {
     assert.equal(cancelled, true);
   });
 }
+
+test('a failed CCTV media fetch reports a fixed health message, not the error text', async (t) => {
+  isolate(t);
+  const root = mkdtempSync(path.join(tmpdir(), 'gev-cctv-'));
+  t.after(() => rmSync(root, { recursive: true, force: true }));
+  mkdirSync(path.join(root, 'config'));
+  writeFileSync(
+    path.join(root, 'config/cctv_sources.austin.json'),
+    JSON.stringify([
+      {
+        id: 'leaky',
+        name: 'Leaky camera',
+        lat: 30.27,
+        lon: -97.74,
+        feedType: 'video',
+        url: 'https://cams.fixture.invalid/leaky.m3u8',
+      },
+    ]),
+  );
+  const leak =
+    'connect ETIMEDOUT 203.0.113.9:443 C:\fixture\secret-path\cams.json';
+  t.mock.restoreAll();
+  t.mock.method(globalThis, 'fetch', () => {
+    throw Error(leak);
+  });
+
+  const call = install(cctvProxy({ sourceRoot: root }));
+  const media = await call('/media/leaky');
+  assert.equal(media.status, 502);
+  assert.equal(media.body.includes('203.0.113.9'), false);
+
+  // GET /health is the second door: src/layers/cctv/frames.js renders each
+  // entry's `message` as the camera's status label, so a raw errno stored here
+  // reaches the screen even though the response above is sanitized.
+  const camera = JSON.parse((await call('/health')).body).cameras.find(
+    (entry) => entry.id === 'leaky',
+  );
+  assert.equal(camera.status, 'degraded');
+  assert.equal(camera.message, 'Media fetch failed');
+  assert.equal(camera.message.includes('ETIMEDOUT'), false);
+  assert.equal(camera.message.includes('secret-path'), false);
+});