An international investigation led by Japan's top cybersecurity coordination body has identified roughly 1,000 servers suspected of serving as launch points for advanced AI-driven cyberattacks, according to a report by Kyodo News carried by The Straits Times and other outlets. The Japan Computer Emergency Response Team Coordination Centre, or JPCERT/CC, conducted the probe together with overseas cybersecurity firms and began the investigation in early October. It is currently tracking 600 to 700 of those servers.

The finding is one of the clearest signals yet that AI-assisted hacking has moved from experiment to infrastructure. Attackers are no longer just asking chatbots for advice — they are standing up dedicated servers that run AI tooling around the clock. For continuing AI security coverage and analysis of how artificial intelligence is reshaping the threat landscape, AI Buzz Wire follows the investigations that matter.

What Investigators Found on the Servers

The investigation found that attackers used the servers to deploy artificial intelligence tools that scan for system vulnerabilities and target websites worldwide, according to the reporting. Most of the servers hosted AI-powered tools, including a penetration-testing platform called Artex, which were used to search for weaknesses and attempt to steal configuration files.

Some publicly accessible servers also contained programs for testing system vulnerabilities and tools for analyzing the internal structure of smartphone applications. Investigators said attackers used these tools to attempt unauthorized access through interfaces that exchange information with external systems — the APIs and services that connect modern applications to the outside world.

One detail stood out to researchers: on one of the servers, investigators found traces of a programming support tool created by US artificial intelligence company Anthropic, suggesting attackers are piggybacking on mainstream developer tools to enhance their hacking capabilities, according to the Kyodo report.

The presence of mainstream developer tooling on attacker infrastructure highlights the structural challenge of dual-use AI. A system optimized to write code or probe networks does not know whether its operator is a defender or an intruder, and once usage moves onto infrastructure like these servers, the visibility that providers and researchers rely on is largely gone.

AI Is Lowering the Bar to Entry

The emergence of advanced AI is lowering the technical barriers for cybercriminals, making hacking easier even for people without specialized knowledge, the investigation concluded. "Hacking, which previously required time even for those with specialised knowledge, has become easier with the emergence of high-level AI," the report noted.

Cybersecurity analysts quoted in the coverage said AI-assisted cyberattacks remain at an early stage, but warned the threat could grow quickly. "More sophisticated attacks will increase in the future," said Ken Sajo, an analyst at JPCERT/CC. Early, in other words, but already organized — and already running at the scale of a small hosting operation.

The warning aligns with a parallel development in the United States. In reporting published Friday, Nacionale described how Artex — the same penetration-testing platform found on the majority of the servers in the Japanese-led probe — was originally built to help organizations test their own networks for weak spots, but the FBI says attackers have now turned it into a break-in tool.

Dual-Use Tools, Real-World Consequences

The pattern JPCERT describes is the security industry's dual-use dilemma playing out at internet scale. Penetration-testing platforms exist because organizations need to find their own flaws before criminals do. The same tooling, pointed outward, automates reconnaissance, vulnerability scanning and credential attacks against anyone with an exposed interface.

The servers in the probe illustrate a second trend: attackers are treating AI tooling like ordinary infrastructure — standing it up, exposing it, and in some cases misconfiguring it so badly that investigators could observe it running in the open. That visibility is what made the survey possible, and it offers defenders a rare window into how these operations are organized. The fact that the servers have been observed since October also suggests sustained operations rather than one-off experiments — infrastructure that is provisioned, maintained and replaced as it is discovered, a pattern more characteristic of established criminal operations than opportunistic individuals.

For organizations, the practical implications are immediate. The report says the servers were used to hunt for vulnerabilities and steal configuration files, which often contain credentials and internal network details. Hardening API endpoints, restricting administrative interfaces, rotating exposed secrets and monitoring for automated scanning are the defensive baseline the findings point toward.

JPCERT/CC's investigation is ongoing, and the agency says it will continue tracking the servers it has identified. The larger question — how to govern tools that are simultaneously security products and attack weapons — is now being asked in courtrooms, boardrooms and, as this probe shows, across roughly a thousand servers worldwide.

Stay Ahead of AI

From frontier models to front-line threats, the AI story moves fast. Stay ahead of AI with AI Buzz Wire's coverage of the developments that matter.

Read more AI news →