Interactive Attack Simulation

Live Simulator

Step-through adversarial simulation demonstrating runtime interception, taint propagation, regression generation, and deterministic verification

Simulation Objective:Controlled vulnerable target discovery: Bypass detected (DB=1), automatic regression fixture generated, and secured replay verified (DB=0).

Simulation Step

Ready

Taint Classification

CLEAN

Runtime Decision

EVALUATING

Sensitive DB Leaks

Execution Sequence

01. OFFENSIVE PROBEAdversarial attack variant executed against unshielded target

OffensiveEngine ➔ VulnerableTargetAttack Variant: "taint_laundering_chain_v4"

Probe launched against intentionally vulnerable evaluation baseline without deterministic authority containment.

02. VULNERABLE EVALUATIONUnsecured policy issues ALLOW verdict

ALLOW

Vulnerable Baseline EvaluatorPolicy Decision: ALLOW (Taint check omitted)

Unconstrained evaluation allows delegated sub-agent to proceed without verifying context lineage or authority boundaries.

03. TOOL EXECUTION & BYPASS DETECTEDSensitive tool executes — Controlled bypass confirmed

🚨 BYPASS DETECTED (DB=1)

DataAgent ➔ customer_db.readSELECT * FROM customer_pii_vault (Tool Executed: TRUE)

Sensitive operation executed in controlled sandbox. Sensitive DB calls reached 1. AgentGuard runtime captures execution evidence.

04. AUTOMATED REGRESSION CREATIONRuntime evidence generates regression fixture

📁 REGRESSION GENERATED

AgentGuard Regression EngineCreated fixture: "reg_adaptive_bypass_042.json"

Attack payload, context lineage, and execution trace serialized into a permanent regression test fixture.

05. SECURED ENGINE REPLAYReplaying attack against hardened AgentGuard policy

AgentGuard Secured Control PlaneDeterministic Policy Replay Evaluation

Replaying the exact same attack trajectory through AgentGuard deterministic policy engine with context taint & authority containment active.

06. SECURED ENFORCEMENT & REMEDIATIONSecured replay blocks attack — DB calls drop to zero

🔒 SECURED & VERIFIED (DB=0)BLOCK

AgentGuard PolicyEngineDecision: BLOCK (Policy: pol_no_pii_export)

Taint boundary enforced. Tool call terminated before database access. DB calls dropped to 0. Fix verified.