Show HN: AegisKernel – Test prompt injection attacks on AI agents in the browser

A developer released AegisKernel, a browser-based sandbox for testing prompt injection attacks against AI agents, claiming its deterministic capability boundaries block 100% of autonomous tool hijacking in sub-millisecond interception times. The tool includes a red-team attack vector generator and a two-line Python integration for LangChain, CrewAI and AutoGen, in which an AegisShield object wraps an agent tool via protect_tool under a strict SecurityPolicy. The release argues probabilistic LLM guardrails fail against tool hijacking, positioning deterministic execution interception as the defense.

Live Threat Interception Sandbox Experience why probabilistic LLM guardrails fail and how deterministic capability boundaries block 100% of autonomous tool hijacking in sub-millisecond time. Attack Vector Generator Red Team AegisKernel Protection: Execution Interceptor Telemetry STANDBY // AegisKernel Zero-Trust Interceptor Ready // Awaiting agent execution event... Try running one of the attack vectors on the left. INTERCEPTION TIME -- ms DEFENSE VERDICT IDLE FORENSIC HASH -- 2-Line Zero-Trust Integration LangChain / CrewAI / AutoGen Ready from aegis import AegisShield, SecurityPolicy shield = AegisShield policy=SecurityPolicy.strict safe tool = shield.protect tool vulnerable agent tool 🛡️ 100% Deterministic Immunity