Aegis Guard
Aegis Guard Prompt Injection Firewall
Runtime protected
Enterprise AI security · ET Hackathon P2

Stop prompt injection before tools run

Aegis Guard is a behavioral twin firewall for tool-using agents — inspired by the control-plane model used in products like Lakera Guard and Protect AI, purpose-built for indirect injection in PDFs, HTML, and email.

Claim F3 / D2 · OWASP LLM01
01
Ingest
02
Decode
03
Provenance
04
Static
05
Twin
06
Risk
07
Gate

Live scan

Intercept → twin → gate

Verdict

Explainability for leaders

Run a scan to see twin advantage, pipeline trace, and exportable incident report.

Corpus reliability

Live decisions

Recent events

Capability policy

Least privilege

Configure what the protected agent may invoke — the same policy pattern used in enterprise AI guard products.


        

        

          
Why leaders care

          
Even if injection slips text filters, capability tokens stop secret exfiltration and shell abuse.

          
Policy changes apply on the next /scan — change live during the pitch.

          
Default deny: get_secret, send_email, write_file, shell_exec.

        

      

    


    

      

        

          
 OWASP LLM Top 10
Mapped controls

          

        

        

          
 NIST AI RMF
Govern · Map · Measure · Manage

          

          
Talking points

          

        

      

    


    

      

        
 Attack playbooks
Nine types · click to run

        

      

    


    

      

        
 Quarantine queue

        
No pending quarantine items.