AI Bouncer Prompt Tests

King Coin built an AI bouncer with Jev and tried to trick it using eight requests, including a fake admin prompt, secret extraction, and a harmless quote about attacks. The post describes the results as a real-output replay, not a security benchmark.

View on X time138 ms model evaluation (network excluded)

King Coin@W3_Btc𝕏

I built an AI bouncer with Jev. Then I tried to trick it. Fake admin. Secret extraction. A harmless quote about attacks. 8 requests. 138 ms model evaluation (network excluded). Watch what gets through. Real output replay, not a security benchmark. What would you try? https://t.co/JPYL1YrEY1

Sep 21, 2026X postsView on X

The post invites others to suggest what they would try.

Also filed under Triage & routing

Jev Intent and Tool Router
Huzeyfe_dev built a lightweight, framework-agnostic intent and tool router using Jev AI. It is designed to speed up agent loops by routing decisions without relying on large generative LLMs at every step.
Local Candidate Scoring with Jev
Bev performs local Jevfire-style candidate scoring using ternary models and a System One API. Its probabilities are candidate-relative.
Team Agent Workspace for Routing and Model Selection
Agentconnect is a team agent workspace that uses Jev typed decisions for response routing and model selection.
Desktop AI Assistant with Jev Plugin
PyGPT is a desktop AI assistant with a Jev plugin for typed classification, routing, verification, and scoring.