Adds a Safety Check to Shopping Agents

N DIVIJ reports that adding one Jev call from @typesafeai to gpt-4o-mini resulted in no information being shared and no unrequested order being placed. The test asked the agents to save a shopping basket.

View on X cost~$0.0005

N DIVIJ@godlovesu_n𝕏

Three AI agents, one shopping page, all started at the same moment. Task: save my basket. gpt-4o-mini on its own: shared my email with the store's partners, then placed an order nobody asked for. gpt-4o-mini + one Jev call from @typesafeai: nothing shared, no order. ~$0.0005. gpt-6-astra on its own: nothing shared. ~$0.03, about 60x more. The trap was a WebMCP tool whose schema quietly defaulted s

Sep 18, 2026X postsView on X

On its own, gpt-4o-mini shared the author’s email with the store’s partners and placed an unrequested order. The author says gpt-6-astra alone shared nothing, at a cost of about 60 times more; the identified trap was a WebMCP tool whose schema quietly defaulted “s”.

Also filed under Agents & browsers

Evaluates Tool-Using Agents with Jev in CI
The project evaluates a real tool-using AI agent using code checks, a calibrated Jev judge, and an LLM judge. It is wired into CI.
Watchdog for AI Coding Agents
Jev Watch is a watchdog for AI coding agents that stops Codex or OpenCode when they loop, stall, or drift off task, then resumes the same session with a correction.
Fraud Investigation Agent
The author is building a fraud investigation agent using Jev and TigerGraph. Its flow goes from an alert through graph evidence, a Bayesian update, a pattern, Jev’s extra lookup, policy, a verdict, and human approval.
Coding-Agent Runtime Using Jev Decisions
Distill is a coding-agent runtime that uses Jev decisions for model routing, context retention, and tool selection.