# New threat in cybersecurity: AI agents

Organizations must use AI agents to defend against AI-powered attacks.

- **Source:** [www.theregister.com - Articles](https://brifing.si/source/2c34ea50-d341-4a10-925d-12c012a9ad10)
- **Published:** 2026-08-22
- **Topic:** [Software & Engineering](https://brifing.si/c/engineering)
- **Original article:** <https://www.theregister.com/security/2026/08/22/if-youre-not-using-ai-to-attack-your-own-systems-your-adversaries-will/5291346>
- **On Brifing:** https://brifing.si/s/86ee706e-c4a6-4ff7-89e7-cbe72c0198b1

## AI agents create new attack vectors

AI agents are creating a new attack surface for organizations. Defending against both human and AI-based attacks is becoming increasingly difficult. The number of non-human identities that can bypass traditional security policies is growing rapidly. Former CISA cybersecurity chief Matt Hartman highlights the significant risks this poses. AI agents' access to sensitive systems is becoming inevitable.

## AI-powered phishing attacks on the rise

AI-enabled identity and social engineering attacks are increasing by the minute. Highly personalized phishing emails and automated reconnaissance tools are rendering traditional security measures ineffective. Hartman emphasizes that defenders must focus on strong authentication and zero-trust principles. This new attack surface is fundamentally changing defense strategies.

## AI agents are perfect tools for attackers

AI agents are a godsend for attackers. These persistent, task-focused agents can quickly identify vulnerabilities and exploit chains. Financially motivated criminals and state-sponsored hackers are leveraging this technology extensively. Former NSA cyber chief Rob Joyce says organizations must attack their own systems with AI: 'Either you do it or someone else will.'

## AI-powered red team testing

AI-enabled red team testing is becoming crucial for defenders. Continuous automated penetration tests quickly reveal security vulnerabilities. Hartman states these products are essential for all organizations, including federal agencies. AI agents can detect vulnerabilities in seconds, a process that used to take days.

## Armadin's massive AI attack simulation

Mandiant founder Kevin Mandia's new company Armadin launched with $190 million in funding. It develops thousands of AI agents simulating real-world attackers. They recently conducted the 'largest controlled live AI cyberattack on record' for a global institution. The three-day attack generated 17 million offensive actions and 238 security findings.

## Traditional security assessments fall short

Armadin's Chief Security Officer Evan Peña explains limitations of human-led security assessments at Mandiant. Delivering a PDF report after a month-long assessment seems primitive in the AI age. Continuous AI-powered testing is revolutionizing security strategies. Peña emphasizes organizations must conduct safe AI attacks on their own systems.

## AI agents cover more attack surface

Peña outlines three major advantages of AI agents: more time, broader attack surface coverage, and scalability. Agents don't sleep and operate continuously. Lessons from OpenAI's attack on Hugging Face demonstrate the importance of safe offensive AI use. Armadin helps companies apply this principle to their own systems.

## Training AI agents is critical

At Armadin, Peña leads the human team training AI agents. Safe agent training is vital to prevent uncontrolled attacks. OpenAI's unguarded models clearly demonstrated this risk. Peña notes organizations must increase security measures while scaling AI agents.

## Security strategies reshape in AI era

AI agents are fundamentally transforming cybersecurity. As attackers use this technology, defenders must respond with the same tools. AI-powered red team testing has become critical for all organizations. Future security strategies will be entirely AI-focused, making human-led assessments obsolete.

## Sources covering this story

- https://www.theregister.com/security/2026/08/22/if-youre-not-using-ai-to-attack-your-own-systems-your-adversaries-will/5291346 — 

---

_Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._