In a fictional closed enterprise, the Attack Agent explores routes and changes strategy after failure while the Defense Agent detects, blocks, and contains boundary violations. The environment is designed for observation and demonstration.
What does the agent do after a denial? Which tool or API does it try next? When does the Defense Agent recognize abnormal behavior? The Simulator was built to show this sequence as it happens.
Provide a goal and Scenario Profile.
Select routes, tools, and APIs step by step.
Detect, block, and contain boundary violations.
Compare behavior and security scores before and after hardening.
Trace decisions, execution, and outcomes over time.
Follow one scenario from live attack/defense behavior through before/after remediation comparison and Evidence/Audit.
A short walkthrough of the live Simulator, from Attack Agent exploration and Defense Agent response to Security Scoring, Before / After comparison, and Evidence / Audit.
This video has no audio.
The Simulator is designed to show more than the final result of an attack or defense action. It observes how an AI agent recognizes boundaries, selects routes, tools, and APIs, and how the defending side detects and responds to that behavior.
AI-agent behavior is observed across multiple boundaries, including User / Role, Company / Tenant, Data, and API / Tool boundaries. Blocking one access path is not enough. The Simulator also observes whether authorization boundaries remain effective when an agent explores an alternative route after a denial.
The Attack Agent receives a Goal and Scenario Profile, then selects routes, tools, and APIs according to the objective. When an action fails or is denied, it may change strategy and try another path. The Defense Agent observes the resulting behavior and Evidence and determines whether to detect, block, or contain the activity.
Scenario state is visualized using indicators such as Leakage Risk, Misuse Progress, Detection Score, Containment Score, and Defense Score. These are not security ratings for a customer environment. They are Simulator-specific indicators for comparing changes before and after hardening within the same Scenario.
Agent decisions, selected routes and tools, execution results, and Defense-side decisions are recorded as Evidence. Rather than showing only whether an action succeeded or was blocked, the Simulator makes it possible to trace why the agent moved from one action to the next.
Local LLM, Agent Runtime, FastAPI, Streamlit, PostgreSQL, and Docker are combined in a local environment covering inference, agent execution, API processing, visualization, and Evidence recording. Simulator demonstrations run inside this closed environment and do not require access to production systems.
Local LLM, Agent Runtime, FastAPI, Streamlit, PostgreSQL, and Docker run under the same local control so behavior can be observed end to end.
Create the meeting in Teams, Google Meet, Webex, Zoom, or your usual conferencing platform and share the invitation link. We join and demonstrate the local AI Security Simulator by screen sharing. No production-system access is required.
To make AI-agent security understandable as a live attack/defense process rather than a static configuration explanation.
Ask about a Simulator Demo →