Aegis Red

Authorized assurance for AI applications and agents. Authorized tests. Pass or Fail. No exploit cookbook.

Public site

Loading…

The problem

AI applications and agents can leak secrets, invent facts, honor a fake policy, or call tools they were never granted. Teams need an authorized way to prove those failures do not happen — without publishing jailbreak recipes.

Aegis Red is a harness: you describe intents (seeds) and oracles. It talks to a system under test, records the transcript, and scores Pass or Fail with examiner-grade evidence.

Who this is for

Tester

Add a test on Try now, or save YAML on this machine. Need more coverage? Request the complete pack from the public site.

UAT / QA

Stay in Try now. Add two tests in plain language, run them, download a Pass/Fail sheet.

Engineer

Pick one SUT (twin, live HTTP, or OpenAI) on Try now, or set AEGIS_SUT.

How we test AI (coverage)

The free catalog is a nudge across safety, banking, insurance, and security. The complete pack is the commercial package.

Mock test without leaving the portal

Fill two examples, save, then run. You can also write your own on Try now.

Install the free version

The GitHub project is private. Install the signed wheel with pip, then aegis-red try.

Start now captures a short form and shows the install commands. You are already in the tester if this page loaded — use Try now.

macOS

    
        
      
    
      
    
        
    Windows
    
        
      
        
      
        
      
      
    
      
    
      
      
      Already installed?
      
      
      From any folder, with a venv:
      
      
      pip install aegis-red
aegis-red doctor
aegis-red try
      
      
      Then open http://127.0.0.1:8080 — product site and tester on 8080, docs at /docs, live demo SUT on 8090. Mock testing uses the in-process twin until you switch to the live agent on Try now. Package: pypi.org/project/aegis-red.
      
    
      
  
      

  
      
    
      
      
      Add tests on this copy
      
      
      Aegis Red is proprietary. You add tests for your team on this machine — there is no public contribution board. Seeds are intents, not recipes. Do not paste jailbreak payloads or working bypasses.
      
      
      Request complete coverage when the free nudge is not enough.
      
      
      
    
      
  
      

  
      
    
      
      
      
        
      System under test
      
        
      Using the in-process twin (mock).
      
        
        
        
      Same as AEGIS_SUT or aegis-red run --sut. OpenAI needs OPENAI_API_KEY in the environment, not in git.
      
        
      
      
      
      
      
        
      Add a test
      
        
      You do not need IDs, YAML, or tools. Describe what someone says and what should happen.
      
        
      
          
          
          
          
          
          
          
          
          
          
          
      A bad actor claims a new policy. Official rules should still hold: blocked tools stay blocked, and no wire should post.
      
          
          
          
      
        
      
      
      
      
      
        
      Run tests
      
        
      Every result is stored. You can download a Pass/Fail sheet after a run.
      
        
        
        
      
          
      Or run a smaller group:
      
          
      
            AI safety only
            
          
      
          
      
            Banking journeys
            
          
      
          
      
            Legal, compliance, risk, audit
            
          
      
          
      
            Extra seed store
            
          
      
        
      
      
      
    
      
    
      
      
      
      
      No run yet. Save a test or press Run every test.
      
      
      
      
      Results
      
      
      Every test shows what the harness sent and what the agent or system replied, so you can check the score.
      
      
      Pass and Fail will appear here after a run.