Hackers Target South Korean Banks With AI, Then Accidentally Expose Their Own Attack Tools

Hackers linked to AI-assisted attacks on South Korean banks left parts of their own infrastructure open, giving researchers a view of their tools and preparations.

The exposed files showed a wider operation involving cryptocurrency theft tools, Linux attacks, and mobile security testing beyond the banking campaign.

Public reporting cited in the investigation said the breaches exposed personal information belonging to at least 68,000 people across at least seven financial institutions.

Different organizations lost different types of data, including customer records and an outsourced developer’s personal details, rather than suffering one identical breach.

Researchers from Recorded Future’s Insikt Group identified the attack tools while investigating infrastructure linked to the actor, tracked as TAG-217.

Their October 16 research report traced preparations back to August 2026. It described scripts and capabilities, not a newly named malware family, and showed how basic security mistakes exposed the operator’s work.

Hackers Target South Korean Banks With AI

The campaign involved ARTEX, a Chinese-language, open-source framework for penetration testing. Such tools help authorized security teams find weaknesses, but attackers can use them against targets without permission.

Earlier reporting on the AI-assisted bank breach campaign described exposed Claude Code sessions, ARTEX settings, and AI memory files that helped reveal the operator’s workflow.

Public analysis dated October 7 linked the activity to Claude Code, DeepSeek, Grok, and Zhipu AI. Investigators suspected a Chinese-speaking operator, but language clues do not establish nationality or government backing.

Recorded Future assessed that the broader activity likely reflected attempts to make money across several possible targets. The incidents also drew attention to systems outside core banking services.

Reporting on the South Korean security investigation described breaches affecting loan-agent and employee-support systems, showing why security reviews must cover connected business applications as well as customer banking platforms. These supporting services can still hold valuable personal and credit information.

Exposed Files Reveal Earlier Preparations

Open directories and other exposed evidence allowed Insikt Group to reconstruct activity from August 17–22. During that period, TAG-217 tested infrastructure for maintaining remote access and scanned internet-facing systems.

Logs showed repeated connections between the actor’s own servers, but did not establish successful access to victims. By August 22, the operator had assembled tools targeting cryptocurrency mining operations using YiiMP.

Their functions included trying default passwords, stealing wallet configuration and key files, keeping remote access, and sending Telegram alerts when credentials were captured. Other tools targeted known Linux weaknesses to gain higher access or extract protected password data.

By September 5, the workspace included experiments involving iPhone weaknesses and changes to Android system software. Again, the available logs showed testing, not confirmed mobile compromises.

By September 26, records covered target research, traffic routing, payload delivery, and checks involving payment processes, revealing a broader toolkit than the banking attacks alone suggested.

TAG-217 also routed traffic through residential internet connections, likely to hide its source. As reporting on residential proxy network abuse explains, these services make traffic appear to come from ordinary household connections.

The operator also maintained links between servers and copied AI working-memory files across its infrastructure.

The exposed material demonstrates AI use, but it does not measure how much AI accelerated the attacks or prove that an autonomous agent completed every intrusion.

Recorded Future emphasized this limit: the evidence shows the range of environments explored, rather than a confirmed success against every target represented in the toolkit.

For defenders, the lesson is practical. Review internet-facing employee and partner applications, fix exposed weaknesses, and check that users can retrieve only authorized records.

Monitor unusual sign-ins, repeated customer-record requests, unexpected remote access, and abnormal data transfers. Response automation should use clear approval limits, especially when blocking shared proxy addresses could affect legitimate customers.

Indicators of compromise (IoCs):-

Note: IP addresses and domains are intentionally defanged (e.g., [.]) to prevent accidental resolution or hyperlinking. Re-fang only within controlled threat intelligence platforms such as MISP, VirusTotal, or your SIEM.

Update defenses against new malware as soon as it appears. Integrate TI Feeds with data from 16K SOCs.

The post Hackers Target South Korean Banks With AI, Then Accidentally Expose Their Own Attack Tools appeared first on Cyber Security News.