Breaking

Unit 42: AI agents helped a human attacker seize enterprise root access in under 10 hours

September 5, 2026 at 04:08 EDT

Security
AI Agents

Palo Alto Networks Unit 42 says a human operator using unnamed frontier AI models and attack-specific agentic frameworks breached an enterprise network and seized root credentials in under 10 hours — work the firm estimates would take human operators around two weeks.

September 2026 · Palo Alto Networks Unit 42

AI agents, human operator, root in under 10 hours

A human attacker used frontier models and agentic frameworks to seize enterprise root credentials in under 10 hours — work Unit 42 estimates would take human operators around two weeks. No zero-day was required.

Time to comparable impact

<10 hours

AI-assisted intrusion

~2 weeks

Human operators (Unit 42 estimate)

Same class of ATT&CK techniques; agents removed the pauses between steps.

50+

MITRE ATT&CK techniques in one loop

Page technical audit left for the victim

Zero-days required, per Unit 42

Attack flow

1. Public web/API breach

→

2. Repo secrets → vault → root

→

3. CI/CD keys, victim AI compute

→

4. Terraform backdoor blocked

What held

Hard branch protection stopped Terraform backdoor injection. Preventive policy did not have to outrun the agents.

What did not

Hard-coded tokens, public-facing access, and hijacked AI endpoints let agents reach root and reuse the victim’s compute. Models used were not named.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.