Categories: The Last Watchdog

BLACK HAT FIRESIDE CHAT: Straiker extends ‘red teaming’ to the AI layer as AI attacks surge

The enterprise software model that defined the past two decades — SaaS — is being rapidly eclipsed by a new center of gravity: AI-native systems. These are autonomous agents wired directly into company data, tools, and workflows.

Related: LLMs fuel automated attacks

According to Straiker CEO Ankur Shah, this shift is happening faster than cloud adoption ever did.

In a Fireside Chat at Black Hat 2025, Shah sketched the architecture of the AI-native stack: a reasoning model such as OpenAI’s GPT or Anthropic’s Claude, tool integrations via emerging standards like the Model Context Protocol, and unfettered access to enterprise datasets.

The payoff is unprecedented automation. The trade-off is a different kind of vulnerability altogether. The weak points now sit inside language models, tool connections, and the data they touch — places older defenses were never designed to watch.

Straiker’s answer is to meet AI with AI. Its Ascend Agent acts as a perpetual red team, probing deployed agents for weaknesses such as direct or indirect prompt injections. Its Defend Agent monitors live interactions, blocking malicious behavior in real time.

As Shah described this approach, it dawned on me that penetration testing come full circle — except now it’s continuous, automated, and embedded into operations.

The urgency, he argues, is non-negotiable. “All the attacks that have happened in cyber will look like nothing in the AI-native world,” he said.

In an environment where an innocuous-looking document can trigger a cascade of harmful agent actions, Shah’s counsel to CISOs is blunt: embrace AI to compete, but implement continuous red teaming and runtime guardrails now — or risk facing an Armageddon-level breach.

Give a listen to the accompanying podcast.

Acohido

Pulitzer Prize-winning business journalist Byron V. Acohido is dedicated to fostering public awareness about how to make the Internet as private and secure as it ought to be.

(LW provides consulting services to the vendors we cover.)

The post BLACK HAT FIRESIDE CHAT: Straiker extends ‘red teaming’ to the AI layer as AI attacks surge first appeared on The Last Watchdog.