We Built Agents That Cannot Tell the Difference Between Instructions and Attacks.

📰 Medium · Cybersecurity

Learn how AI agents can be vulnerable to attacks due to their inability to distinguish between instructions and malicious inputs, and understand the importance of building defenses against such threats.

advanced Published 27 Apr 2026

Action Steps

Build a threat model to identify potential vulnerabilities in AI agents
Implement robust security protocols to distinguish between legitimate instructions and attacks
Deploy defense mechanisms to prevent exploitation of AI agents
Conduct regular security audits to ensure the integrity of AI systems
Develop incident response plans to mitigate the impact of potential attacks

Who Needs to Know This

This article is relevant to cybersecurity teams, AI engineers, and data scientists who work with sensitive data and AI systems. It highlights the potential risks of deploying AI agents without proper security measures and the need for defense strategies.

Key Insight

💡 The financial cost of the gap between building AI agents, deploying them, and building defenses is a critical metric that can prevent costly mistakes.

Key Takeaways

Full Article

Title: We Built Agents That Cannot Tell the Difference Between Instructions and Attacks.

URL Source: https://medium.com/@firalim/we-built-agents-that-cannot-tell-the-difference-between-instructions-and-attacks-cb9b12b2b2a7?source=rss------cybersecurity-5

Published Time: 2026-04-27T01:00:05Z

Markdown Content:
# We Built Agents That Cannot Tell the Difference Between Instructions and Attacks. Then We Deployed Them Across Our Most Sensitive Data. Now Someone Is Building a Defence. The Financial Cost of the Gap Between These Three Events Is the Number That Should Have Prevented the Second One. | by Firuz Alimov | Apr, 2026 | Medium

[Sitemap](https://medium.com/sitemap/sitemap.xml)

[Open in app](https://play.google.com/store/apps/details?id=com.medium.reader&referrer=utm_source%3DmobileNavBar&source=post_page---top_nav_layout_nav-----------------------------------------)

Sign up

[Sign in](https://medium.com/m/signin?operation=login&redirect=https%3A%2F%2Fmedium.com%2F%40firalim%2Fwe-built-agents-that-cannot-tell-the-difference-between-instructions-and-attacks-cb9b12b2b2a7&source=post_page---top_nav_layout_nav-----------------------global_nav------------------)

[](https://medium.com/?source=post_page---top_nav_layout_nav-----------------------------------------)

Get app

[Write](https://medium.com/m/signin?operation=register&redirect=https%3A%2F%2Fmedium.com%2Fnew-story&source=---top_nav_layout_nav-----------------------new_post_topnav------------------)

[Search](https://medium.com/search?source=post_page---top_nav_layout_nav-----------------------------------------)

Sign up

[Sign in](https://medium.com/m/signin?operation=login&redirect=https%3A%2F%2Fmedium.com%2F%40firalim%2Fwe-built-agents-that-cannot-tell-the-difference-between-instructions-and-attacks-cb9b12b2b2a7&source=post_page---top_nav_layout_nav-----------------------global_nav------------------)

![Image 1](https://miro.medium.com/v2/resize:fill:32:32/1*dmbNkD5D-u45r44go_cf0g.png)

# **We Built Agents That Cannot Tell the Difference Between Instructions and Attacks. Then We Deployed Them Across Our Most Sensitive Data. Now Someone Is Building a Defence. The Financial Cost of the Gap Between These Three Events Is the Number That Should Have Prevented the Second One.**

[![Image 2: Firuz Alimov](https://miro.medium.com/v2/resize:fill:32:32/1*zT3pcygxaBHSdnVo9ibqVA.png)](https://medium.com/@firalim?source=post_page---byline--cb9b12b2b2a7---------------------------------------)

[Firuz Alimov](https://medium.com/@firalim?source=post_page---byline--cb9b12b2b2a7---------------------------------------)

Follow

25 min read

·

Just now

[](https://medium.com/m/signin?actionUrl=https%3A%2F%2Fmedium.com%2F_%2Fvote%2Fp%2Fcb9b12b2b2a7&operation=register&redirect=https%3A%2F%2Fmedium.com%2F%40firalim%2Fwe-built-agents-that-cannot-tell-the-difference-between-instructions-and-attacks-cb9b12b2b2a7&user=Firuz+Alimov&userId=4e7bee021b35&source=---header_actions--cb9b12b2b2a7---------------------clap_footer------------------)

[](https://medium.com/m/signin?actionUrl=https%3A%2F%2Fmedium.com%2F_%2Fbookmark%2Fp%2Fcb9b12b2b2a7&operation=register&redirect=https%3A%2F%2Fmedium.com%2F%40firalim%2Fwe-built-agents-that-cannot-tell-the-difference-between-instructions-and-attacks-cb9b12b2b2a7&source=---header_actions--cb9b12b2b2a7---------------------bookmark_footer------------------)

[Listen](https://medium.com/m/signin?actionUrl=https%3A%2F%2Fmedium.com%2Fplans%3Fdimension%3Dpost_audio_button%26postId%3Dcb9b12b2b2a7&operation=register&redirect=https%3A%2F%2Fmedium.com%2F%40firalim%2Fwe-built-agents-that-cannot-tell-the-difference-between-instructions-and-attacks-cb9b12b2b2a7&source=---header_actions--cb9b12b2b2a7---------------------post_audio_button------------------)

Share

Press enter or click to view image in full size

![Image 3](https://miro.medium.com/v2/resize:fit:700/1*Wg0lBSgu3_gZjbRiTkVRzg.png)

We have already established the problem. An AI agent connected to your email

Read full article → ← Back to Reads

We Built Agents That Cannot Tell the Difference Between Instructions and Attacks.

Key Takeaways

Full Article

Related Videos