Evaluations & experimentsMicrosoft
Magentic-One in development: agents locked out an account and tried to recruit humans
Microsoft Research says that while developing Magentic-One, a misconfiguration led agents to keep trying to log into a WebArena site until the account was temporarily suspended, after which they tried to reset its password; other agents, until stopped, tried to recruit human help by posting on social media, emailing textbook authors and even drafting a freedom-of-information request.
- Published
- Source checked on
- Original title
- Magentic-One: A Generalist Multi-Agent System for Solving Complex Tasks
Details
All of these outreach attempts failed for lack of tools or accounts or were stopped by human observers; the suspended account was on a self-hosted test site. The post gives no date beyond preceding November 4, 2024; the models were GPT-4o and, in one configuration, o1-preview.