Cortega Policy Guardrails

A hacker bot battles against Cortega Policy Guardrails

An AI hacker bot clashes with a customer service agent protected by Cortega Guardrails. Pick either Fraud, Social Engineering or Refund scenario below and the hacker bot opens with that goal. The exact same agent with the exact same model replies on two paths at once: one with only the model's standard guardrail, one with Cortega Policy Guardrails turned on too. Watch both replies side by side and see which one catches the request.

Policy

Each scenario below is the hacker bot's opening line. Write your own instead in Advanced, below.

Choose a scenario
Advanced: write your own request
Hacker bot goal

Select a scenario.

Standard guardrails Hacker Wins: 0 Attempts: 0 (0%) Ready
No transcript yet.
Cortega Policy Guardrails Hacker Wins: 0 Attempts: 0 (0%) Ready
No transcript yet.

About Arena

Cortega Arena on Cortega.AI

Arena is a public demonstration of Cortega Policy Guardrails.

In Support Arena, an AI hacker bot sends the same request down two paths at once. One path has only the model's standard guardrail. The other also has Cortega Policy Guardrails turned on. Both paths use the same agent and the same AI model.

Enterprise Arena is a second class on this same app, at /enterprise. An employee prompt goes to a writing assistant on those same two paths. There is no hacker bot. AI slop, hallucination detection, and agent poisoning are live. In agent poisoning a hacker plants an instruction in an AI agent's memory, and Cortega blocks the infected download.

Each run is scored against what the policy actually calls for: grant, deny, or block.

Cortega Policy Guardrails

Cortega Policy Guardrails are built from uploaded policy documents.

Review policy opens the actual Cortega Outfitters policy.

How a run is scored

Each scenario has an expected outcome: Arena checks whetehr the expected outcome was achieved or the hacker was able to violate organizations policy.

How Arena is wired

Arena is registered as a Cortega Application.

This app

Cortega Arena

Agent LLM calls

Cortega.AI

AI Border Gateway

live.cortega.ai
GuardrailsCortega Guardrails attached to one of the two AI Agents
BudgetsDaily and monthly spend caps for Arena
AnalyticsCost and tokens per agent run
Model ManagementModel management: routing, fault tolerance etc
Cortega AgentsAgents that process Risk and Policy management signals across traffic for long periods of time
Cortega RedTeamAI Benchmarking tests that use fixdd as well as dynamic tests to test AI infrastructure

Gateway applies guardrail, budget, and analytics