diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 7738b34..b62e4e8 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -712,7 +712,8 @@ jobs:
               "Accessibility=suite-artifacts/accessibility.json" \
               "Visual Regression=suite-artifacts/reports/visual.json" \
               "Performance=suite-artifacts/performance.json" \
-              "Load=suite-artifacts/load.json"
+              "Load=suite-artifacts/load.json" \
+              || echo "TESTS_FAILED=1" >> "$GITHUB_ENV"
             echo ""
             echo "**Platform:** ${{ env.BASE_URL }}"
             echo ""
@@ -739,3 +740,11 @@ jobs:
             reports/e2e_combined.json
             reports/e2e_summary.md
           retention-days: 30
+
+      # Every suite's pytest step is `continue-on-error` so all reports get
+      # produced; this is where a failing test finally turns the run red.
+      - name: Fail on test failures
+        if: env.TESTS_FAILED == '1'
+        run: |
+          echo "::error::One or more suites have failing tests — see the results table in the job summary."
+          exit 1
diff --git a/scripts/suite_summary.py b/scripts/suite_summary.py
index 4b5d13a..5047a0f 100644
--- a/scripts/suite_summary.py
+++ b/scripts/suite_summary.py
@@ -19,7 +19,8 @@
 
 Each argument is `Label=path`. Missing or unparseable files are reported as
 such rather than skipped silently — a suite whose report never arrived is a
-result worth seeing, not a blank row. Shard globs are supported so the E2E
+result worth seeing, not a blank row. Exits 1 when any suite has failing
+tests, so CI can gate on it. Shard globs are supported so the E2E
 matrix can be passed as one label:
 
     python scripts/suite_summary.py "E2E=reports/e2e_shard_*.json"
@@ -112,9 +113,8 @@ def main(argv: list[str]) -> int:
     print("\n".join(rows))
     print()
     if any_failure:
-        print("> One or more suites have failing tests. Job status alone does not")
-        print("> reflect this — every suite runs with `continue-on-error: true`.")
-    return 0
+        print("> One or more suites have failing tests.")
+    return 1 if any_failure else 0
 
 
 if __name__ == "__main__":