Inside the Agentic Red Team — Fighting AI With AI

Why It Matters

Continuous AI‑driven red‑team testing gives enterprises real‑time visibility into emerging AI‑related attack surfaces, turning costly, periodic assessments into an ongoing defensive advantage.

Key Takeaways

• Sweet Security launches “Sweet Attack,” an AI‑driven continuous red‑team service.
• Uses models like Mythos and GPT‑5.5 to simulate AI‑powered attacks.
• Provides real‑time vulnerability recipes and automated remediation guidance.
• Shifts focus from periodic pen‑testing to ongoing AI‑based defense.
• Aims to prioritize and mitigate AI‑generated code vulnerabilities at scale.

Summary

Techstrong TV introduced Sweet Security’s newest offering, Sweet Attack, a fully automated red‑team platform that leverages cutting‑edge large language models to probe cloud environments for AI‑related weaknesses. The service replaces traditional, infrequent pen‑testing with a continuous, AI‑powered scan that can generate exploit “recipes,” flag misconfigurations in real time, and suggest remediation steps.

The product relies on models comparable to Anthropic’s Mythos and OpenAI’s GPT‑5.5, turning what used to be a costly, manual engagement into a scalable, token‑driven process. Sweet Security emphasizes that while token costs may approach human‑team expenses, the 24/7 coverage and ability to simulate AI‑driven attackers provide unmatched breadth and speed. Customers receive actionable payloads they can test themselves, turning abstract reports into concrete proof‑of‑concept attacks.

During the interview, Yugal Alberger highlighted the shift from “vulnerability discovery” to “vulnerability defense,” noting that the industry is moving toward governance, prioritization, and rapid remediation. He described the approach as “fighting AI with AI,” stressing that as more code is generated by AI, continuous AI‑based red‑team testing becomes essential to keep pace with emerging threats.

The launch signals a broader transformation in cloud security: enterprises must adopt continuous, model‑driven defenses to protect increasingly AI‑centric workloads. By automating red‑team exercises, Sweet Attack promises faster threat identification, better prioritization of fixes, and ultimately a higher baseline of secure code as AI‑generated software proliferates.

Original Description

In this episode of TechStrong TV, Alan Shimel sits down with Yigael Berger, Chief AI Officer at Sweet Security, to talk about how the new generation of code-aware AI models is rewriting the rules of offensive and defensive security — and to break the news on Sweet Attack, Sweet Security's new agentic AI red team service, now generally available.

Yigael explains how Sweet's runtime-first cloud security platform now extends into proactive red teaming, why a once-or-twice-a-year red team engagement is no longer enough, and how Sweet Attack packages what used to be a high-end consulting service into a continuous, AI-driven product. The conversation also dives into the so-called "vulnerability apocalypse," the shift from vulnerability discovery to prioritization and remediation, and what happens when AI is both writing and attacking code.

Chapters:

00:00 Introduction

00:35 Yigael's background and role as Chief AI Officer

02:00 What Sweet Security does — runtime-first cloud security

04:30 Announcing Sweet Attack — agentic AI red team service

07:00 Token cost vs. human pen testers and 24/7 scale

09:00 Continuous red teaming for dynamic AI environments

11:00 From vulnerability discovery to defense and prioritization

13:30 AI writing the code and AI defending the code

15:30 Closing thoughts

Learn more: https://sweet.security

Subscribe to TechStrong TV for more conversations with the leaders shaping DevOps, cybersecurity, AI, and cloud-native technology.

#AgenticAI #RedTeam #CloudSecurity #AIForSecurity #TechStrongTV #SweetSecurity #PenTesting #DevSecOps #VulnerabilityManagement #AppSec