Unbelievable! Chinese hackers use AI to siphon South Korean bank customers’ personal data, then expose themselves by asking Claude to “write a résumé”

動區 BlockTempo · 2026-10-09T02:04:51.000Z

Artificial intelligence (AI) has turned out to be a double-edged sword that led a hacker to expose himself! A new report from leading international cybersecurity firm CrowdStrike revealed that a hacker suspected of being Chinese used the open-source AI penetration tool ARTEX and Anthropic’s Claude model to successfully breach several South Korean financial institutions and leak customers’ personal data. But the hacker’s server misconfiguration exposed his Claude chat logs. Along with asking where to sell South Korean personal data, he even asked AI to write a cybersecurity researcher résumé containing his name, phone number, university, and Telegram account—leaving security experts both amused and exasperated. (Previously: South Korean banks hit by consecutive cyberattacks; government points to traces of a “Chinese-language AI hacking tool”) (Background: A white-hat hacker used Claude to access OpenAI’s internal code repository, but received a bounty of only $6,500.) Generative AI is playing an increasingly complex role in cybersecurity, but now it has become the unlikely star of an absurd tale of a hacker exposing himself. On October 7, 2026, international cybersecurity giant CrowdStrike published a new investigative report stating that the perpetrators behind the large-scale cyberattacks on South Korean financial institutions, which began in late September, used the Chinese-developed open-source AI agent tool ARTEX alongside large language models such as DeepSeek v4.1-flash and Anthropic’s Claude Code to automate their attacks. Yet despite the sophistication of the hacker’s methods, a basic security lapse left his conversations with AI completely exposed in a public directory. In a spur-of-the-moment request for AI to “write a résumé,” he inadvertently revealed clues to his real identity to the entire internet. ### AI agents automate intrusions, exposing the personal data of tens of thousands of South Korean customers The attack campaign targeting South Korea’s financial system began in late September. According to statements from the affected banks, Shinhan Bank confirmed that the personal data of approximately 25,000 customers had been accessed without authorization, while the personal information of 119 customers at Kookmin Bank was also exposed. The targets included a loan application status inquiry system and an employee mobile-office support platform. CrowdStrike’s investigation found that the attackers primarily used ARTEX, a Chinese-developed open-source AI penetration-testing tool (an agentic AI penetration tool), to carry out the attacks. Its technical setup connected DeepSeek v4.1-flash as the primary backend and also called on Zhipu AI’s GLM-5.3, xAI’s Grok 4.6, and Anthropic’s Claude Code, greatly increasing the speed and efficiency of automated vulnerability detection and attacks. ### Asked Claude where to find buyers—and to help create a “cybersecurity résumé” Although the attack combined cutting-edge AI technology, the hacker made a critical mistake in securing his infrastructure. In an open directory on a server controlled by the attacker, CrowdStrike found complete Claude Code session logs, configuration files, and memory files, making it possible to reconstruct the attacker’s actions in detail. The chat logs showed that the hacker asked Claude where he could sell South Korean personal data and to search for South Korean data-trading groups on Telegram. In another conversation, he asked Claude to help write a résumé for a cybersecurity researcher and to include his “achievements” from using ARTEX to attack South Korean banks. The résumé prompt inadvertently revealed a trove of personal information: Name/alias: Y.Y Age and date of birth: Listed as 26 years old (the initial date of birth given was 2007-09-22) Education: South China University of Technology Location: Maoming, Guangdong Province, China Contact details: Mobile number 17820191556, Telegram account @YY520CN ### Identity clues laid bare, signaling new cybersecurity threats in the AI era The accidentally exposed résumé quickly sent shockwaves through the industry. According to South Korean media reports, a journalist tried calling the mobile number listed in the report. A man in Henan who identified himself as a convenience store clerk answered and denied any connection to the incident. It remains unclear whether the résumé used the real personal information of someone else or a fabricated identity. South Korean President Lee Jae-myung expressed serious concern about the incident during a cabinet meeting, noting that AI’s involvement in cybercrime has caused significant public anxiety. South Korean police have formally established a task force to conduct a comprehensive investigation. CrowdStrike assessed that, while it could not confirm the attacker’s ultimate identity with complete certainty, it had moderate confidence that the perpetrator was a Chinese-speaking individual with financial motives. The firm warned that AI agent tools lower the barrier to sophisticated attacks, and that threat actors will increasingly use AI to accelerate the pace of their operations. Businesses and financial institutions must urgently upgrade their cybersecurity defenses. Related coverage Codex + DeepSeek: Hackers use AI agents to breach systems at 395 facilities in 48 countries The next move for crypto hackers: No more billion-dollar bets—will they use AI agents to steal small amounts every day? AI aids crime! Hackers use Anthropic’s Claude to easily breach the Mexican government and steal 150 GB of sensitive data A job interview turns out to be a North Korean hacker trap! PurpleBravo breaches more than 3,100 IP addresses, with AI and cryptocurrency companies as prime targets Unbelievable! Chinese hackers use AI to freely siphon South Korean bank customers’ personal data, then expose themselves by asking Claude to “write a résumé” This article was first published on BlockTempo, the most influential blockchain news outlet.

人工智慧（AI）竟成為駭客自爆的雙面刃！國際頂尖資安機構 CrowdStrike 最新報告揭露，一名疑為中國籍的駭客利用開源 AI 滲透工具「ARTEX」與 Anthropic 的 Claude 模型，成功入侵韓國多家金融機構並外洩客戶個資。然而，這名駭客卻因伺服器設置疏失，外洩了 Claude 的對話紀錄 —— 其中除了詢問「去哪賣韓國個資」外，甚至還要求 AI 幫他撰寫包含姓名、電話、畢業學校與 Telegram 帳號的「資安研究員履歷」，讓資安專家啼笑皆非。 （前情提要：韓國銀行業連續被駭，政府指出有「中文 AI 駭客工具」痕跡） （背景補充：白帽駭客靠 Claude 摸進 OpenAI 內部程式碼庫，賞金僅獲 6,500 美元）   生成式 AI 在資安攻防戰中扮演的角色日益複雜，如今卻演變成一場令人啼笑皆非的「自爆」劇碼。國際資安大廠 CrowdStrike 於 2026 年 10 月 7 日發布最新調查報告指出，韓國金融機構自今年 9 月下旬起遭遇的大規模網路攻擊，背後幕後黑手使用了中國開發的開源 AI 代理工具「ARTEX」，搭配 DeepSeek v4.1-flash 與 Anthropic 的 Claude Code 等大型語言模型進行自動化攻擊。 然而，這名手法精密的駭客卻因為低級的安全疏失，將其與 AI 的對話紀錄完全暴露在公開目錄中，更因一時興起要求 AI「幫寫個人履歷」，意外向全網公開了自身的真實身分線索。 AI 代理自動化入侵，韓國數萬名客戶個資外洩 這起針對韓國金融體系的攻擊活動始於 9 月下旬。根據受害銀行發布的聲明，新韓銀行（Shinhan Bank）證實約有 25,000 名客戶的個人資料遭非法存取，國民銀行（Kookmin Bank）亦有 119 名客戶個資外洩。駭客攻擊標的涵蓋了貸款進度查詢系統以及員工行動辦公支援平台。 CrowdStrike 的追蹤顯示，攻擊者主要透過中國開發的開源 AI 滲透測試工具 ARTEX（代理式 AI 滲透工具）發動襲擊。其技術架構串接了 DeepSeek v4.1-flash 做為主要後端，同時還交叉調用了智譜 AI 的 GLM-5.3、xAI 的 Grok 4.6 以及 Anthropic 的 Claude Code，大幅提高了自動化偵測漏洞與攻擊的節奏與效率。 叫 Claude 尋找買家，還要求 AI 幫忙打造「資安履歷」 儘管攻擊手法結合了最前沿的 AI 技術，但駭客在基礎設施的防護上卻出現了致命漏洞。CrowdStrike 在攻擊者控制的伺服器開放目錄中，發現了完整的 Claude Code 工作階段對話紀錄、設定檔與記憶檔，得以完整還原攻擊者的操作過程。 對話紀錄顯示，該名駭客除了請 Claude 尋找「在哪裡可以販售韓國個資」以及「搜尋 Telegram 上的韓國資料交易群組」外，還在一段對話中要求 Claude 協助撰寫一份資安研究員的個人履歷，並要求將本次利用 ARTEX 攻擊韓國銀行的「戰果」納入履歷經驗中。這段履歷提示詞意外揭露了大量個人資訊： 姓名代號：Y.Y 年齡與出生日期：標記 26 歲（初給出生日期為 2007-09-22） 學歷背景：華南理工大學 地理位置：中國廣東省茂名市 聯絡方式：手機號碼 17820191556、Telegram 帳號 @YY520CN 身分線索全曝光，預示 AI 時代下的新型資安威脅 這份無意間公開的履歷迅速引發業界震撼。據韓媒報導，記者嘗試撥打報告中的手機號碼，一名位於河南、自稱便利商店店員的男子接聽後否認與此事有關，目前尚無法確定該履歷是否為駭客使用他人真實資訊或虛構身分。 韓國總統李在明在國務會議上特別針對此事件表達高度關切，指出 AI 參與網路犯罪已對社會公眾造成顯著不安，韓國警方已正式成立專案小組展開全面調查。 CrowdStrike 評估認為，雖然無法百分之百確認攻擊者個人的最終歸屬，但以中等信心判斷幕後黑手為具備財務動機的中文使用者。該機構警告，AI 代理工具降低了高階攻擊的門檻，未來威脅者必將更頻繁地利用 AI 加速攻擊節奏，企業與金融機構必須加速升級資安防禦體系。 相關報導 Codex + DeepSeek 駭客用 AI 代理攻陷 48 國、395 家設施系統 加密駭客下一步：不再豪賭十億，改用 AI 代理天天偷小錢？ AI 助攻犯罪！駭客靠 Anthropic Claude 輕鬆入侵墨西哥政府，偷走 150GB 敏感資料 求職面試竟是北韓駭客陷阱！PurpleBravo入侵逾3,100個IP地址，AI與加密貨幣公司成最大獵物〈傻眼！中國駭客靠 AI 爽洗韓國銀行客戶個資，竟因叫 Claude「幫寫履歷」自爆身分〉這篇文章最早發佈於動區BlockTempo《動區動趨-最具影響力的區塊鏈新聞媒體》。