AI-Assisted Ransomware Gang Tore Through a Corporate Network in Under 10 Hours
Researchers at Palo Alto Networks watched attackers use artificial intelligence agents to do in a single working day what normally takes criminal crews two weeks. The case is a signal, not an outlier.
Key points
- Palo Alto Networks' Unit 42 research team documented a ransomware intrusion completed in under 10 hours that analysts estimated would take human operators roughly two weeks without AI assistance.
- The attackers used AI agents, meaning software programs that act independently and adjust their own next steps based on what they find, to automate reconnaissance, credential theft, and system access across the victim's network.
- The criminals broke in through a public-facing API endpoint, a doorway that lets outside software talk to internal systems, before spreading to cloud services, source-code repositories, and a secrets-management vault holding administrator passwords.
- Existing branch protections, a setting that requires human approval before code changes go live, blocked the attackers' attempt to plant hidden backdoors inside the victim's infrastructure-management files.
- Analysts say the core risk is not new attack techniques but a sharp drop in the time, skill, and cost needed to run existing ones at speed.
A ransomware gang used artificial intelligence to tear through a corporate network in fewer than ten hours, according to Unit 42, the threat-research division of Palo Alto Networks. Ransomware is malicious software that locks a company's files and systems until a payment is made. The researchers estimated that equivalent work, done by human operators alone, would have taken around two weeks.
The group has not been publicly named. What Unit 42 documented, first reported by CSO Online, is a playbook that borrowed familiar criminal techniques but handed the execution to AI agents, software programs that read the results of each step and decide the next move without waiting for a human to issue the instruction.
How did the attackers get in?
The criminals entered through a public-facing API endpoint, a connection point that lets external software communicate with a company's internal systems. From there, an automated reconnaissance agent mapped the internal network, identifying services the way a burglar would case each room of a building.
Separate agents then searched the company's source-code repositories, the libraries where developers store their work, looking for exposed credentials: usernames and passwords left in the code by mistake. Those credentials opened a secrets-management system, a secure vault where companies store administrator passwords and access keys. With vault access, the attackers obtained full administrative control.
| Stage | What happened |
|---|---|
| Initial access | Broke in through public-facing API endpoint |
| Reconnaissance | AI agent mapped internal network services |
| Credential theft | Agents found exposed passwords in code repositories |
| Privilege escalation | Gained admin access via secrets-management vault |
| Cloud pivot | Used stolen cloud keys to access victim's AI computing resources |
| Attempted persistence | Tried to plant backdoors in Terraform config files; blocked |
The attackers then hijacked a code-management application to steal cloud access keys, which are digital passes that grant access to a company's rented computing resources hosted online. They turned the victim's own cloud computing power into infrastructure for further attacks.
They also attempted to plant backdoors inside Terraform configuration files. Terraform is software companies use to build and manage cloud infrastructure through written instructions rather than manual clicks. A backdoor hidden there would survive a cleanup. Branch protections, a standard setting requiring a human to approve code changes before they take effect, blocked the attempt.
The attackers confirmed to Unit 42 researchers during ransom negotiations that they had used frontier AI models and purpose-built attack frameworks. Whether the attack was fully autonomous is unclear. Analysts who reviewed the case described it as a human-directed intrusion where AI handled the tactical, step-by-step work.
Should businesses be worried about the speed?
Yes, because most security teams are not set up to respond in hours. Standard detection-and-containment processes were built for intrusions that unfold over days.
Analysts advising on the case recommend that organisations move away from long-lived credentials, passwords and keys that stay valid for months, toward short-lived credentials that expire quickly and limit the damage if stolen. Security providers should also be authorised in advance to disable a breached account immediately, without waiting for internal approval, so the window between detection and containment shrinks.
Detecting this kind of attack also requires joining the dots across systems that are often watched separately. A suspicious action in one environment looked unremarkable alone; its significance only became clear when connected to activity elsewhere on the network.
What should ordinary employees watch for?
This particular intrusion targeted technical systems rather than individual inboxes, but stolen credentials remain the most common entry point across ransomware cases broadly. Employees who reuse passwords across work and personal accounts, or who store passwords in shared documents or code files, create the kind of exposure the attackers exploited here.
If your company notifies you of a security incident, change your work passwords promptly and watch for unusual account activity or unexpected password-reset emails you did not request.
