Close Menu
The Financial News 247The Financial News 247
  • Home
  • News
  • Business
  • Finance
  • Companies
  • Investing
  • Markets
  • Lifestyle
  • Tech
  • More
    • Opinion
    • Climate
    • Web Stories
    • Spotlight
    • Press Release
What's On
The Matrix Of Leadership Toy Review: You Got The Touch!

The Matrix Of Leadership Toy Review: You Got The Touch!

June 18, 2026
Iran’s President Signs Interim Peace Deal After Trump—Key Details Of Agreement Shared

Iran’s President Signs Interim Peace Deal After Trump—Key Details Of Agreement Shared

June 18, 2026
Best Prompt Engineering Techniques And Strategies From A To Z

Best Prompt Engineering Techniques And Strategies From A To Z

June 18, 2026
Why Is Iran Signing The US Deal? Hormuz And Sanctions

Why Is Iran Signing The US Deal? Hormuz And Sanctions

June 18, 2026
Today’s NYT Mini Hints And Answers For Thursday, June 18

Today’s NYT Mini Hints And Answers For Thursday, June 18

June 18, 2026
Facebook X (Twitter) Instagram
The Financial News 247The Financial News 247
Demo
  • Home
  • News
  • Business
  • Finance
  • Companies
  • Investing
  • Markets
  • Lifestyle
  • Tech
  • More
    • Opinion
    • Climate
    • Web Stories
    • Spotlight
    • Press Release
The Financial News 247The Financial News 247
Home » The Importance Of Red Teaming For Scaling Enterprise AI Agents

The Importance Of Red Teaming For Scaling Enterprise AI Agents

By News RoomMay 22, 2026No Comments5 Mins Read
Facebook Twitter Pinterest LinkedIn WhatsApp Telegram Reddit Email Tumblr
The Importance Of Red Teaming For Scaling Enterprise AI Agents
Share
Facebook Twitter LinkedIn Pinterest Email

Joan Vendrell, NeuralTrust CEO and cofounder, has 15+ years of technology leadership experience advancing enterprise-grade AI security.

​I recently spoke with a CISO who was preparing for a major production rollout of an autonomous customer service agent. They had passed their traditional penetration tests with flying colors. But when I asked how the agent would handle a multi-step prompt injection attack that evolved in real time, there was a long silence. “We tested the model last month,” they finally said. “But the agent is learning and interacting with live data every hour.”

This is the fundamental challenge of the agentic era. Traditional security testing is a snapshot in time, while agentic AI is a continuous movie. At a time when agents are being granted the authority to execute workflows, call APIs and access sensitive databases, relying on a “one-and-done” security audit is like checking the locks on a house while the walls are still being built.

We are seeing a shift where the attack surface is not just the code or the network, but the reasoning process itself. If we don’t move toward a model of continuous red teaming, we aren’t just leaving the door open; we are handing the keys to the house to an autonomous operator we haven’t fully vetted.

The Problem: The Dynamic Attack Surface And “Adversarial Reasoning”

The core issue is that AI agents are non-deterministic. Unlike a standard application where input A always leads to output B, an agent’s behavior changes based on its context, its memory and the tools it has access to. This creates a playground for what I call “adversarial reasoning”: attacks designed to corrupt the agent’s logic rather than just its input.

Gartner predicts that by 2028, more than 50% of enterprises will use dedicated AI security platforms to manage these risks. The reason is simple: the OWASP Top 10 for LLM Applications has evolved. We aren’t just worried about simple prompt injections. We are now facing “agentic hijacking” and “indirect prompt injection,” where an agent is manipulated through the very data it is supposed to analyze.

In my experience, traditional red teaming—where a human team spends two weeks trying to break a system—cannot keep up with the speed of AI development. We need a “machine-versus-machine” approach to security.

5 Steps To Implementing Continuous Red Teaming

To secure the agentic enterprise, we must move beyond static testing and embrace a proactive, continuous defense. Here are five steps I believe every security leader should take.

​1. Automate the adversary with attacker agents.

If your agents are operating 24/7, your red teaming must do the same. You need to deploy adversarial agents with the sole job of finding weaknesses in your production agents.

This is about stress-testing the agent’s reasoning. Can it be tricked into bypassing a safety guardrail? Can it be convinced to escalate its own privileges? By using the MITRE ATLAS framework to map these attacks, you can automate the discovery of vulnerabilities before a malicious actor does.

​2. Stress-test the tool-use and API boundaries​.

The most dangerous part of an AI agent isn’t the model, but the tools it can call. Red teaming must focus heavily on “insecure output handling,” a top risk in the OWASP 2025 list.

You need to simulate scenarios where an agent is given a malicious command through a trusted tool, such as a compromised email or a poisoned database entry. Can an agent be tricked into executing a “delete” command on a database because it “reasoned” it was the right thing to do? Testing these boundaries is the new frontier of security.

​3. Align with the NIST AI risk management framework.

​The NIST AI RMF provides a critical taxonomy for managing AI risk. In my view, continuous red teaming should be mapped directly to the NIST “Measure” and “Manage” functions. This ensures that your testing isn’t just a series of random attacks, but a structured validation of your risk tolerance.

By using a standardized framework, you can provide the board with measurable data on your AI security posture, moving from “we think we’re safe” to “we know we’re resilient.”

​4. Simulate indirect prompt injection scenarios.

​One of the most insidious threats today is indirect injection. Imagine an agent reading a public website to summarize news, only to find a hidden instruction in the HTML that tells it to exfiltrate the user’s session cookie. Your red team must continuously feed your agents “poisoned” data to see if they can maintain their instructions.

In my experience, the best defense is a zero-trust approach to agent inputs. Never assume the data an agent retrieves is safe.

​5. Focus on “identity lineage” during attacks.

​During a red team exercise, pay close attention to how the agent’s identity is used. Does the agent maintain a clear chain of accountability when it’s under pressure? If an adversarial prompt tricks an agent into performing an unauthorized action, can you still trace that action back to the original human intent?

Continuous red teaming should validate that your identity lineage remains unbroken, even when the agent’s reasoning is compromised.

The Bottom Line: Security Is A Living Process

The rise of agentic AI is the most significant shift in enterprise technology in a generation, but it requires a new level of discipline. We cannot secure autonomous systems with manual, point-in-time processes.

In my experience, the companies that will lead the next decade are those that treat security not as a hurdle to be cleared, but as a living, breathing process. By embracing continuous red teaming, we don’t just find vulnerabilities; we build the resilience necessary to let our AI agents move faster, do more and transform our businesses with confidence.​​​

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?

Joan Vendrell
Share. Facebook Twitter Pinterest LinkedIn Tumblr Email

Related News

The Matrix Of Leadership Toy Review: You Got The Touch!

The Matrix Of Leadership Toy Review: You Got The Touch!

June 18, 2026
Best Prompt Engineering Techniques And Strategies From A To Z

Best Prompt Engineering Techniques And Strategies From A To Z

June 18, 2026
Today’s NYT Mini Hints And Answers For Thursday, June 18

Today’s NYT Mini Hints And Answers For Thursday, June 18

June 18, 2026
The Scientist Who Risked Everything To Understand Sharks

The Scientist Who Risked Everything To Understand Sharks

June 18, 2026
Justin Gaethje Gets Suspension After Beating Topuria

Justin Gaethje Gets Suspension After Beating Topuria

June 18, 2026
How To Download The iOS 27 Developer Beta For Free On Your iPhone Now

How To Download The iOS 27 Developer Beta For Free On Your iPhone Now

June 17, 2026
Add A Comment
Leave A Reply Cancel Reply

Don't Miss
Iran’s President Signs Interim Peace Deal After Trump—Key Details Of Agreement Shared

Iran’s President Signs Interim Peace Deal After Trump—Key Details Of Agreement Shared

News June 18, 2026

ToplineIran’s President Masoud Pezeshkian signed the interim peace agreement early on Thursday, hours after President…

Best Prompt Engineering Techniques And Strategies From A To Z

Best Prompt Engineering Techniques And Strategies From A To Z

June 18, 2026
Why Is Iran Signing The US Deal? Hormuz And Sanctions

Why Is Iran Signing The US Deal? Hormuz And Sanctions

June 18, 2026
Today’s NYT Mini Hints And Answers For Thursday, June 18

Today’s NYT Mini Hints And Answers For Thursday, June 18

June 18, 2026
Stay In Touch
  • Facebook
  • Twitter
  • Pinterest
  • Instagram
  • YouTube
  • Vimeo
Our Picks
2026 U.S. Open At Shinnecock Hills. .5 Million Prize Purse At Stake

2026 U.S. Open At Shinnecock Hills. $22.5 Million Prize Purse At Stake

June 18, 2026
The Scientist Who Risked Everything To Understand Sharks

The Scientist Who Risked Everything To Understand Sharks

June 18, 2026
AEW Dynamite Results For June 17, 2026 As Forbidden Door Nears

AEW Dynamite Results For June 17, 2026 As Forbidden Door Nears

June 18, 2026
Justin Gaethje Gets Suspension After Beating Topuria

Justin Gaethje Gets Suspension After Beating Topuria

June 18, 2026
The Financial News 247
Facebook X (Twitter) Instagram Pinterest
  • Privacy Policy
  • Terms of use
  • Advertise
  • Contact us
© 2026 The Financial 247. All Rights Reserved.

Type above and press Enter to search. Press Esc to cancel.