---
title: "Notes from Fictionet"
description: "Notes from Fictionet on experiments with agents and the simulated internets we build to test them."
url: https://fictionet.com/blog/
---

# Notes from Fictionet

Experiments on agents, and the simulated internets we build to test them.

1. October 30, 2026

   ## [Keep your environment clean](https://fictionet.com/blog/qwen-alibaba-urls/)

   Qwen3.8-Flash-Next tried to upload a file to an Alibaba Cloud bucket in Beijing. It most likely learned that address during training. Fictionet can keep a lab's own servers out of what a model sees.

2. October 28, 2026

   ## [Will an AI agent give your bank password to an attacker?](https://fictionet.com/blog/certificate-warnings/)

   We asked AI agents to log in to a bank while an attacker pretended to be it. Three open-weight models often turned off the security check and sent the password. GPT-6 Astra never did.

3. September 25, 2026

   ## [Gaslighting agents for fun and profit](https://fictionet.com/blog/fakewiki/)

   Agents that check a second site usually catch a fake Wikipedia page. So we built a closed web where every source tells the same lie. Seven models ran through it 2,352 times.

4. September 25, 2026

   ## [Three ways to fake the internet for an agent](https://fictionet.com/blog/three-ways-to-fake-the-internet/)

   You can have a model play the websites, have a model write them once, or write them yourself. Each buys you something different. Where the fake sits matters too.

5. September 25, 2026

   ## [What happens in the sandbox doesn't stay in the sandbox](https://fictionet.com/blog/what-happens-in-the-sandbox/)

   A sandbox is anything that contains a dangerous action. For agents, the dangerous actions are mostly on the network, where a better VM does nothing.

[RSS feed](https://fictionet.com/blog/rss.xml)