What happens when cybercriminals no longer need a team of hackers to launch sophisticated attacks against multiple businesses simultaneously?
A newly documented cyberattack campaign provides a troubling answer.
On October 27, 2026, cybersecurity researchers at TrendAI revealed that attackers had used CyberStrikeAI, an offensive security platform coordinating 44 AI agents, to automate activities ranging from vulnerability discovery to exploitation, persistence, and data theft.
The findings highlight an important change in cybersecurity: attackers can now automate much of the work that previously required significant human effort.
And that raises an uncomfortable question for business leaders.
If attackers can operate at machine speed, can a cybersecurity strategy built primarily around detecting and responding to threats keep up?
Key Takeaway: AI is making cyberattacks faster, more scalable, and less dependent on human operators. However, the fundamental actions attackers need to perform have not disappeared. Businesses should strengthen prevention by restricting unauthorized activity, rather than relying exclusively on detecting threats after they begin.
Researchers discovered three separate attack infrastructures targeting organizations in Japan, each demonstrating a different level of AI involvement.
According to the original TrendAI research, the most automated operation used CyberStrikeAI to coordinate 44 specialized AI agents.
These agents divided responsibilities across the attack process, including:
In a relatively short period, the platform conducted validation activities against at least 19 identifiable organizations, including 16 in Japan.
Importantly, researchers confirmed that only a small proportion showed evidence of deep compromise. The 19 organizations should not be interpreted as 19 successful breaches.
The researchers also documented a separate human-led campaign involving AI assistance that scanned at least 6,000 Japanese IP addresses. Another operation evaluated at least 1,300 Oracle WebLogic candidates and 35 VPN entry points.
These were distinct operations, not necessarily the work of the same attackers.
Because the cost and effort required to conduct cyberattacks are declining while the potential scale is increasing.
Consider the traditional attack model. A criminal identifies a target, researches its systems, tests vulnerabilities, attempts to gain access, and decides what to do next.
Each step requires time, expertise, and coordination.
Now imagine software agents handling many of those tasks simultaneously, around the clock.
Instead of pursuing one potential victim at a time, an attacker can investigate hundreds or thousands of systems and concentrate on those presenting the easiest opportunities.
For businesses, the consequences can include compromised credentials, stolen customer information, operational disruption, recovery expenses, and reputational damage.
The concern is not simply that AI makes attackers smarter.
It allows attackers to do more, faster, with fewer people.
Not in the cases examined by TrendAI. The researchers found that attackers largely exploited familiar vulnerabilities, exposed administration interfaces, default credentials, and misconfigured systems.
One of the clearest CyberStrikeAI intrusion chains began with an unauthenticated administration interface on a legacy web application.
Another operation gained access to a geographic information system through default credentials, exposing business records, Windows credentials, and email information.
These weaknesses existed before agentic AI.
What changed was the ability to locate, test, and exploit them repeatedly with much less human intervention.
That distinction matters.
Changing the attack does not necessarily change the endpoint actions the attacker ultimately needs in order to succeed.
Whether an intrusion is orchestrated by a human or 44 AI agents, attackers may still need to execute processes, abuse legitimate applications, access sensitive files, steal credentials, establish persistence, or move through the network.
Those activities create opportunities for prevention.
Detection and response technologies remain essential, but AI-driven attacks increase the pressure on systems that depend on recognizing suspicious activity and responding before damage occurs.
Endpoint Detection and Response (EDR) solutions provide valuable visibility, investigation, and automated response capabilities.
However, several challenges become more significant when attacks are automated.
An AI-driven operation can distribute activity across numerous targets, adjust techniques, and use legitimate administrative tools that are difficult to distinguish from authorized activity.
TrendAI specifically observed that the CyberStrikeAI platform throttled its activity against individual targets, making the broader campaign harder to recognize from any one organization's logs.
This does not mean EDR is ineffective or that these particular attacks successfully bypassed EDR.
It means businesses must consider what happens when malicious activity is not detected quickly enough.
Detection can help identify an attack. Prevention controls can also restrict what an attack is allowed to accomplish.
Yes, certain malicious actions can be prevented without first identifying the specific malware, AI agent, or attack technique responsible.
This is where Isolation and Containment become particularly relevant.
Instead of allowing applications broad freedom to operate and relying entirely on security tools to recognize malicious behavior, Isolation and Containment establish boundaries around what applications and processes are permitted to do.
These restrictions can limit unauthorized execution, access to sensitive resources, system modification, and other behaviors attackers may need to complete their objectives.
Consider an AI agent that generates an entirely new malicious script.
Traditional detection systems may need to analyze the script's behavior or recognize indicators of compromise.
A properly configured prevention control may instead restrict the script from launching unauthorized processes or manipulating protected files, regardless of whether the script has ever been seen before.
What if you did not have to detect the attack in order to stop it?
That is the principle behind AppGuard, a proven endpoint protection solution with more than a decade of production history focused on prevention through Isolation and Containment.
AppGuard restricts unauthorized Windows endpoint behavior without requiring the attack to first be identified as malicious.
It complements detection and response rather than replacing those capabilities.
There is an important limitation: endpoint containment alone cannot correct an exposed web application, revoke stolen cloud credentials, or prevent every type of data theft. Those risks require additional security controls.
Nevertheless, restricting what attackers can execute and access after reaching a Windows endpoint can reduce their opportunities to cause damage.
Unknown does not automatically mean unstoppable.
Businesses should prepare for a threat environment where attackers can discover and exploit weaknesses faster than traditional security processes may accommodate.
The objective is not to predict every attack. It is to restrict the actions an attacker needs in order to succeed.
CyberStrikeAI demonstrates how artificial intelligence is transforming the economics and operational scale of cybercrime.
Forty-four agents can coordinate attack activities that previously required substantial human effort.
Yet the underlying research also reinforces something important: many successful intrusions continue to depend on familiar weaknesses and predictable attacker behaviors.
Businesses should not abandon detection and response. They should recognize that detecting every attack before damage occurs is an increasingly difficult objective.
A more resilient security strategy combines visibility and response with prevention controls that limit what an attacker can accomplish.
The attack may be new. The actions it needs to perform often are not.
For additional perspective, read our related article: AI Agents Are Changing Cyberattacks. Can Your Defenses Keep Up?
TrendAI: Not Only ARTEX, The Different Levels AI Is Used in Attacks Targeting Japan
The Hacker News: CyberStrikeAI Uses 44 AI Agents to Automate Hacking Operations