A Chinese-developed AI penetration-testing tool called ARTEX has been found on infrastructure used to attack Japanese companies, raising concerns about how legitimate security tools can be repurposed for cybercrime.
Japanese cybersecurity firm GMO Cybersecurity by Ierae disclosed the findings on October 19. Its investigation identified ARTEX on one attack infrastructure targeting approximately 20 Japanese companies, followed by the discovery of six additional AI-enabled attack servers. Eight targeted companies had publicly disclosed data breaches during September and October.
The findings follow CrowdStrike's investigation into ARTEX use against South Korean financial institutions. However, investigators have not established that the same attackers operated in both countries.
ARTEX found on infrastructure targeting Japanese companies
ARTEX is an AI-powered penetration-testing system designed to identify vulnerabilities and assist with security assessments. Its capabilities can also be misused to automate parts of an intrusion.
GMO's investigation identified one attack infrastructure targeting approximately 20 Japanese companies, including nine websites from the same industry.
The researchers subsequently discovered six additional AI-enabled attack servers primarily targeting Japanese organizations. The industries included insurance, finance, logistics, e-commerce, retail, and education.
Importantly, GMO specifically identified ARTEX on the initial infrastructure. Its findings do not establish that ARTEX operated on every additional server.
The investigation also identified eight targeted companies that had publicly disclosed data breaches in September and October. GMO believes multiple incidents involved the same attack group and shared its findings with law enforcement.
These findings expand the regional concerns already raised by AI-linked attacks on South Korean financial institutions.
Neither the tool's Chinese origin nor its presence on attack infrastructure establishes that the operators were Chinese or acting on behalf of a government.
JPCERT/CC documents AI-assisted API attacks
A separate investigation by Japan's national incident-response organization, JPCERT/CC, identified AI-assisted attacks against Japanese organizations and services intended for users outside Japan.
The researchers confirmed that attackers used AI to scan for known vulnerabilities and conduct attacks through application programming interfaces (APIs).
One technique involved analyzing Android application packages to identify APIs and test them for vulnerabilities, including server-side request forgery (SSRF).
Investigators found files associated with Blutter and jadx, tools used to reverse-engineer Flutter and Android applications. Other evidence included scripts for SSRF testing.
On a separate server not directly linked to the investigated attacks, JPCERT/CC found evidence of attempted endpoint detection evasion and web shells believed to have been deployed inside a targeted network.
JPCERT/CC did not identify ARTEX in its report or establish that its findings involved the same operators investigated by GMO.
The agency also cited independent security research demonstrating how an embedded RSA private key and application-signing logic could be extracted from a Flutter app to reproduce authenticated API requests. That demonstration was not presented as a confirmed step in the Japanese attacks.
Security measures for organizations
The attack techniques documented by JPCERT/CC place particular pressure on organizations that expose APIs through mobile applications.
Its recommendations include:
- Protect application secrets: Never embed private keys or long-lived credentials in mobile applications.
- Enforce server-side authorization: Validate user identities and permissions independently of application-level signing or obfuscation.
- Monitor API activity: Apply rate limiting and investigate unusual request patterns.
- Restrict SSRF exposure: Limit external destinations accessible to APIs and validate redirects.
- Investigate suspicious infrastructure: Review relevant logs and indicators when investigating possible unauthorized access.
These precautions complement broader defenses against AI-assisted cyberattacks, particularly where automated tools can repeatedly probe exposed services.
What remains unconfirmed
GMO's findings establish ARTEX use on attack infrastructure targeting Japanese companies, but the identities of the operators remain unknown.
CrowdStrike separately documented ARTEX use in attacks against South Korean financial institutions. Its investigation identified AI agent configuration files and session histories, but did not attribute the campaign to a named threat actor.
The two investigations establish use of the same tool in separate regional attacks, not a common operator.
For Japanese organizations, the immediate priority is examining exposed applications and APIs for weaknesses attackers can exploit, regardless of whether those attacks involve ARTEX or another automated tool.
Also read: The US Army's AI hacking initiative raises additional questions about controlling autonomous offensive security tools.