Author(s): Ahmed Boulahia

Originally published on Towards AI.

Prompt injection can manipulate your AI agent into leaking data or behaving unpredictably. What is it exactly, and how to beat it?

In this current AI-dominated era of programming and web development, there is an ever-growing inclination towards integrating LLMs through Chatbots and Agents into web and software products. However, like any other new technology in its early days, it is prone to malicious attacks.

Chatbots and Agents aren’t an exception to this rule. There are several types of malicious attacks that can target LLM based applications in 2025, as reported by The Open Worldwide Application Security Project (OWASP), and “prompt injection” is at the top of the list.

Recently, I have came across a tweet by @jobergum about a Github repository that includes all the system prompts of famous production-level agents like Cursor, Windsurf, Devin, etc..

This can be achieved via a meticulously designed attacks such as Jailbreaking and prompt injection, and it clearly shows us that even production-level LLM systems are vulnerable to attacks, and without robust measures to counter such attacks companies risk not only compromising user trust and data security but also losing valuable clients, and suffering significant financial losses.

In this post, I will explain to you what is prompt injection, how it is used maliciously to target LLM-based applications, and the different ways to defend against it.

The chatbots and agents… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI