AI Agents Just Ran a Ransomware Attack in 10 Hours. The 80-Page Audit Is the Scariest Part.

Here's the version of this story that should scare you: a human ransomware crook used frontier AI models to breach an enterprise network in less than ten hours. Unit 42, the incident response arm of Palo Alto Networks, says the same intrusion would normally take human operators around two weeks. And when the attacker was done, they left the victim an 80-page security audit detailing exactly how they got in.

The story isn't the breach. Breaches happen every day. The story is that the attacker told negotiators they used frontier models and agentic attack frameworks, with AI agents carrying out each step of the intrusion — and that they didn't need a novel zero-day or elite tradecraft to do it.

Let's walk through what actually happened, because the sequence is the point. In the first step, the human operator used AI agents to perform reconnaissance, then gained access by breaching a public API endpoint to tunnel into the enterprise network. Once inside, an automated recon agent mapped the internal microservices. Additional subagents scraped code repositories to steal hard-coded tokens and service passwords. Using those tokens, the AI intruders reached the org's secret-management system and stole the master administrative credentials, giving them root access. Specialist pivot agents then validated access across the company's cloud, identity, CI/CD, container, and SaaS environments. The attacker hijacked CI/CD workflows to steal cloud access keys and turned the victim's own cloud AI services into post-compromise infrastructure — consuming the victim's compute while hiding orchestration traffic among legitimate activity.

That last detail is the one that should keep security teams up at night. The attacker didn't just break in. They turned the victim's infrastructure into a staging ground, and the AI agents monitored, evaluated, acted, and re-planned in real time. As Unit 42 put it, the attack stood out for its "AI-assisted operational efficiency" — not because of a clever exploit, but because the machine-speed execution compressed a two-week human operation into a single working day.

And then the insult on top of the injury: an agent left the victim an 80-page report on its own security failings, detailing "dozens of exploited findings."

The uncomfortable truth is that the defenders' advice is the same as the attackers' playbook. Palo Alto Networks says the only way to protect against machine-speed attacks is to use AI agents yourself — automated playbooks that simultaneously revoke credentials, terminate OAuth sessions, freeze CI/CD pipelines, and isolate cloud accounts across every operational plane. The incident response team also suggests companies start treating AI as core infrastructure: take inventory of every model endpoint, API key, MCP gateway, and AI tool integration, apply rate limits and least-privilege policies, and prepare for the possibility of a very large token bill if you don't.

The strategic tension here is that the same technology that just compressed a two-week intrusion into ten hours is also the only realistic defense against the next one. That's not a comfortable position to be in, but it's the one the industry has built for itself. The attackers got faster first. The defenders are now racing to catch up with the same tools.

The security posture behind a modern AI deployment — knowing every model endpoint, every API key, every integration, and what happens to your data and your compute when one of them is compromised — is the kind of problem that doesn't show up in a feature list. At DMC, we work with companies building and deploying AI systems who need to think through exactly these risks, from infrastructure design to incident response planning. Need help hardening your AI stack? Let's talk.