AI agent security
Applications that put an LLM in the loop have their own weaknesses: prompt injection, tool abuse, data leakage. This is a recent area for SIFNOS, approached from the practice of development.
This page takes stock: what exists, what is in preparation.
In development
Conversational agent analysis tool
A tool that stress-tests bots and agents. Not published yet.
In preparation
Vulnerability research
First research will go through coordinated disclosure: maintainers are warned and their fix is awaited before publication.
In preparation
Bug bounty programmes
Account opened on a bug bounty platform; first reports to come.
Ongoing
Training
Self-training in AI agent security: reading, exercises on training platforms, building a vocabulary and a method.
Publications will appear on the Research page. To report a vulnerability in this site, see the legal notice.