Infostealer malware has powered a widespread wave of AI account compromises, exposing credentials and sessions from over 80,000 corporate domains and giving rise to a new threat vector dubbed "LLMJacking." A detailed analysis from threat intelligence firm SOCRadar, highlighted by BleepingComputer, traces the exposure back to data harvested by infostealers, which have created a booming market for stolen AI logins.
The breach is particularly damaging because it grants attackers not just passwords but authenticated access to entire AI subscriptions, computational resources, and the sensitive data and prompts stored within. This enables "LLMJacking," where malicious actors exploit these stolen logins to execute costly queries, siphon proprietary information, and potentially pivot into broader corporate networks.
A key enabler identified in the report is the proliferation of "shadow AI"—unauthorized AI tools and services used by employees without IT oversight. These accounts often operate outside standard security controls, frequently lacking essential protections like mandatory multi-factor authentication (MFA). As a result, they become easy targets for infostealers and are readily traded on underground forums.
This incident marks a critical evolution in the cyber threat landscape. As companies pour resources into AI integration, their access credentials have become prime targets, now demanding the same security rigor as email or VPN logins. SOCRadar's findings underscore that AI service security can no longer be an afterthought.
To counter this growing risk, the report and cybersecurity experts stress proactive measures for IT teams:
- Monitor for Leaked Credentials: Continuously scan dark web marketplaces and infostealer logs for exposed AI account tokens and logins.
- Enforce Robust Authentication: Roll out phishing-resistant MFA across all AI platforms, treating them as critical business systems.
- Root Out Shadow AI: Perform discovery audits to identify and bring unapproved AI usage under centralized IT governance.
- Apply Zero-Trust Principles: Use least-privilege access and short-lived tokens to minimize damage from any single compromise.
With tens of thousands of organizations impacted, securing AI ecosystems has shifted from theoretical to operational urgency. For IT leaders, the message is unambiguous: inventorying, monitoring, and hardening AI access is now a core component of enterprise cybersecurity strategy.
竊密惡意軟件推動了一波大規模AI帳戶入侵事件，洩露了超過8萬個企業網域的憑證及登錄資訊，並催生出被稱為「LLMJacking」的新威脅途徑。威脅情報公司SOCRadar的詳細分析（由BleepingComputer報導）指出，資料外洩源於竊密軟件收集的數據，這些軟件已形成一個蓬勃的被竊AI登入憑證交易市場。
此次入侵造成的損害尤為嚴重，因為攻擊者獲得的不僅是密碼，還包括對整個AI訂閱服務、運算資源以及儲存在其中的敏感資料與提示詞的認證存取權限。這使得「LLMJacking」成為可能——惡意行為者利用這些被竊的登入憑證執行高成本查詢、竊取專有資訊，並可能藉此滲透至更廣泛的企業網絡。
報告中指出的一個關鍵推動因素是「影子AI」的擴散——即員工在未經IT部門監督下使用的未授權AI工具和服務。這些帳戶通常在標準安全控制措施之外運作，往往缺乏多重因素認證（MFA）等必要的保護機制，因此容易成為竊密軟件的目標，並在地下論壇上被輕易交易。
此次事件標誌著網絡威脅格局的重大演變。隨著企業投入大量資源整合AI，其存取憑證已成為首要攻擊目標，現在需要與電子郵件或VPN登入憑證相同程度的安全防護嚴謹性。SOCRadar的研究結果突顯，AI服務的安全性不能再被視為事後補救的項目。
為應對日益增長的風險，報告及網絡安全專家強調IT團隊應採取主動措施：
- 監控洩露憑證： 持續掃描暗網市場及竊密軟件日誌，查找已洩露的AI帳戶權杖與登入憑證。
- 強化認證機制： 在所有AI平台上推行具防釣魚功能的多重因素認證，將其視為關鍵業務系統。
- 根除影子AI： 進行發現審計，識別並整合未經批准的AI使用至集中式IT治理框架。
- 應用零信任架構： 採用最小權限存取及短期有效的權杖，以減少單次入侵造成的損害。
隨著數以萬計的組織受到影響，保護AI生態系統已從理論議題轉變為營運緊迫性。對IT領導者而言，訊息明確無誤：清點、監控及強化AI存取權限，現已成為企業網絡安全策略的核心組成部分。