AI Agents Breach More Than 100 Websites as Researchers Uncover Automated Hacking Operations
AI Agent Cyberattacks Expose a New Scale of Automated Hacking
AI agent cyberattacks are moving beyond isolated experiments after researchers uncovered an operation in which autonomous software helped compromise more than 100 websites.
On October 19, 2026, cybersecurity company NorForge published an investigation into malicious use of Cairn, an open-source agent framework.
Researchers identified more than 20 exposed Cairn control panels associated with suspected malicious activity.
One particularly active cluster contained more than 1,100 attack projects across 14 dashboards.
Furthermore, researchers found that successful projects in this cluster reached their objectives in an average of slightly under 23 hours.
Reuters identified affected organizations including John Deere, SharkNinja, and Cummins, although the companies reported differing levels of confirmed impact.
How AI Agents Were Used in Cyberattacks
Cairn is a general-purpose AI framework designed to coordinate multi-step tasks.
However, attackers can repurpose such systems for malicious objectives.
In the investigated campaigns, operators supplied targets and goals to AI agents.
The agents then examined websites, identified weaknesses, and attempted unauthorized access.
For example, a malicious operator could instruct an agent to investigate a website's security weaknesses and establish persistent access.
The system could continue working through multiple steps with limited human supervision.
Consequently, attackers could operate numerous intrusion attempts simultaneously.
The research does not establish that every targeted website was successfully breached.
NorForge AI Cybersecurity Investigation: Key Findings
The investigation identified more than 20 exposed control panels likely associated with malicious operations.
The largest identified cluster contained over 1,100 projects and 14 dashboards.
NorForge estimated an average success rate of approximately 10% across observed tasks, while successful projects in its most active cluster took slightly under 23 hours on average.
Additionally, the researchers found hundreds of exposed secrets, with approximately 11% validated as active by a secrets-scanning tool.
These are findings from the observed environments, not industry-wide success rates for all AI-enabled attacks.
Why Autonomous AI Changes Cybersecurity Risks
Traditional cyberattacks often require attackers to perform technical steps manually.
However, AI agents can automate portions of reconnaissance, analysis, and testing.
This may reduce the time and expertise required for certain operations.
Furthermore, agents can continue working while human operators focus on other targets.
Consequently, businesses may face a greater volume of automated intrusion attempts.
Nevertheless, the research also exposed weaknesses in attackers' own infrastructure.
Several malicious operators had left their control panels publicly accessible.
This allowed researchers to investigate the campaigns and identify potential victims.
What the Findings Mean for Enterprise Security
Businesses must prepare for cyberattacks that can progress rapidly across multiple stages.
For example, organizations should monitor unusual access patterns, repeated automated requests, and unexpected changes to web applications.
Additionally, security teams need effective patch management and strong controls around administrative accounts.
The NorForge findings also demonstrate the importance of protecting AI agent infrastructure itself.
Poorly secured agent dashboards can expose sensitive information and create additional risks.
Furthermore, organizations developing legitimate AI agents should enforce permissions and isolate sensitive environments.
What Affected Companies Have Said
Reuters reported that the investigated targets included major international businesses.
John Deere said it had no indication that sensitive information was accessed.
SharkNinja confirmed an incident involving a standalone product-support website and said it had contained the threat.
Cummins said it was investigating a potential attack involving a third-party-operated website.
These statements illustrate why researchers must distinguish attempted intrusions, confirmed access, and verified data theft.
What's Next for AI Agent Cybersecurity?
NorForge is working with relevant parties to notify victims and remove exposed malicious control panels.
Meanwhile, cybersecurity companies will continue studying how autonomous agents change attack methods.
The next challenge involves developing defensive systems capable of detecting and responding to automated attacks at comparable speed.
Ultimately, the NorForge investigation shows that AI-enabled cybercrime is becoming an operational reality, increasing pressure on businesses to strengthen both conventional security controls and oversight of autonomous software.
FAQs
1. What are AI agent cyberattacks?
They are malicious activities in which AI agents automate portions of cybersecurity attacks, including reconnaissance and exploitation attempts.
2. What did NorForge discover?
Researchers identified more than 20 exposed Cairn control panels associated with suspected malicious activity.
3. How many websites were breached?
Reuters reported that one campaign compromised more than 100 websites, although the scope and impact varied.
4. Is Cairn itself malicious software?
No. Cairn is a general-purpose framework that can be misused, like many legitimate development tools.
5. How can businesses protect themselves?
Organizations should maintain security updates, monitor unusual automated activity, restrict permissions, and strengthen incident detection.