Runtime Protection Agent
Web Application & API Protection
On-demand Webinar
On-demand Webinar

Your Agent Has Gone Rogue: Dissecting Emerging Attacks Against DevSecOps AI Tools | Harness Resource | Harness

Watch our OnDemand webinar to explore the revolutionary advancements Agentic AI offers DevSecOps, including secure coding, runtime security, and threat detection. This webinar, featuring cutting-edge research from ASPEN Labs, delves into popular AI-powered security use cases like real-time threat scoring, dynamic policy generation, and automated code fixes that make engineering and security teams more efficient. However, these powerful AI capabilities introduce new vulnerabilities that attackers can exploit through evasion, manipulation, and abuse.

Watch this webinar to:

  • Understand AI Attack Methods: Learn about attack patterns that target AI, like context window poisoning, prompt injection, token smuggling, and function calling abuse;
  • Discover Prompt Engineering Impacts: See how prompt engineering significantly influences vulnerability likelihood and exploitability, even in leading AI models;
  • Re-think Hardening of AI-native Security Tools: Gain insights into attack vectors to secure your Agentic AI configurations against emerging and advanced threats.

‍

Published
November 20, 2025

Guide on its way

Check your inbox — your playbook is ready.

View Now!

You're all set

Check your inbox — your download is on the way.

Redirect link

Web Application & API Protection

Redirect link

What you'll learn

Key Takeaways

Rapid AI Development Outpaces Security Measures

The speed of shipping code using AI has doubled, while security efforts have struggled to keep pace. This imbalance results in a high volume of unreviewed and potentially insecure applications.

Non-Developers Introduce New Application Vulnerabilities

Agentic AI allows individuals without coding experience to build and deploy applications. This shift often leads to the creation of inherently insecure software.

Connected Agent Swarms Multiply Security Risks

Linking multiple probabilistic AI agents together into automated workflows increases the likelihood of compounding errors. A mistake by a single agent can cascade through the entire system.

AI Security Tools Are Vulnerable To Injection

DevSecOps tools and firewalls that leverage AI are now susceptible to prompt injection attacks. Attackers can manipulate these systems to bypass security guardrails or embed malicious backdoors.

AI Policy Generation Can Produce Incompatible Code

When using AI for dynamic policy generation, large language models sometimes create incompatible regular expressions. This can cause security engines to fail or process rules incorrectly.