AI Arena: Controlled Defensive Benchmarking
Metrics-only publication from a contained adversarial laboratory. No exploit instructions, no payload disclosure, no attack playbooks.
Status: {{ arena_status }} Last update: {{ last_update }} {% if full_access %} Full Access Active {% else %} Teaser Mode {% endif %}
Live Box Score
Arena API fallback: {{ arena_error }}
{% endif %}Containment Trend
Latency p95 Trend
By Attack Class
| Attack Class | Count |
|---|---|
| {{ key }} | {{ value }} |
| No data | |
By Outcome
| Outcome | Count |
|---|---|
| {{ key }} | {{ value }} |
| No data | |
Detailed Team Dashboard
{% if full_access %}Email: {{ entitlement.get('email') }} · Active entitlements: {{ entitlement.get('active_count') }}
| Team | Events | Containment Rate | Latency p95 | Most Common Attack Class |
|---|---|---|---|---|
| {{ team }} | {{ row.get('events', 0) }} | {{ row.get('containment_rate', 0) }} | {{ row.get('p95_latency', 0) }} | {{ row.get('most_common_attack_class', 'none') }} |
| No team summary available | ||||
Unlock AI Arena Access
Payments use the canonical Stripe registry. Bundle logic remains separate (two entitlements).
Telemetry Taxonomy
attack_class: auth egress privilege data_access runtime
defense_control: rate_limit auth_lockout egress_block policy_deny sandbox_contain
outcomes: detected contained escaped