According to a study by threat intelligence firm SOCRadar, corporate credentials for generative AI platforms are emerging as prime targets for infostealer malware, with researchers identifying 482 companies exposed in stealer logs over a 90-day period.
The findings, reported by Security Affairs, indicate a significant shift in cybercriminal tactics. Attackers are now actively harvesting corporate AI account sessions, API keys, and authentication tokens, moving beyond traditional logins to target the credentials that grant access to powerful, cloud-based AI services.
SOCRadar's examination of the data found that 295 of the 482 affected companies appeared in active logs during the monitoring period, suggesting these exposures are fresh and likely being actively exploited or traded. The research highlights why these assets have become so valuable: a single compromised enterprise AI credential can offer a concentrated entry point with extensive permissions.
Unlike a standard email login, a stolen key for a service like Microsoft Copilot or an Azure OpenAI deployment can lead to immediate financial damage through fraudulent compute usage. Furthermore, it can expose sensitive proprietary logic and internal prompts, and provide a pivot point into other connected corporate systems and data stores.
This threat flourishes in the governance gap created by rapid, often decentralized, AI adoption. As departments scramble to deploy AI tools for productivity gains, the credentials and API keys powering these services are frequently managed with less rigor than core SaaS applications. This creates a "shadow AI" security blind spot, where sensitive keys may be stored insecurely, logged in plaintext, or shared carelessly, making them vulnerable to the endpoint-harvesting techniques of infostealer malware.
The criminal underground has taken notice. SOCRadar notes that stolen credentials for popular AI platforms are now commanding high premiums on illicit forums, reflecting strong demand for this new class of access credential.
For organizations, especially those leveraging cloud AI for competitive advantage, the findings represent a critical call to action. Securing AI assets must be integrated into standard security protocols. Key measures include:
- Audit and Inventory: Discover all sanctioned and unsanctioned AI service subscriptions, API keys, and service accounts.
- Centralize Governance: Manage AI credentials with the same rigor as database passwords, utilizing dedicated secrets management solutions.
- Enforce Conditional Access: Apply zero-trust principles, including mandatory multi-factor authentication and restrictions to managed devices.
- Monitor for Anomalies: Watch for unusual usage patterns on AI platforms, such as massive data queries or atypical login locations.
- Rotate Credentials: Implement regular policies for the rotation of API keys and service account secrets.
The convergence of infostealers and corporate AI credentials marks a pivotal evolution in the threat landscape. As AI becomes embedded in business-critical workflows, securing its access pathways is no longer optional; it is a fundamental operational necessity.
根據威脅情報公司 SOCRadar 的一項研究，生成式 AI 平台的企業憑證正成為竊取惡意軟件的主要目標，研究人員在為期 90 天的日誌中識別出 482 家受影響的公司。
據《Security Affairs》報導，這些發現揭示了網絡犯罪策略的重大轉變。攻擊者現在正積極竊取企業 AI 帳戶的會話、API 密鑰和認證代幣，目標已超越傳統登入憑證，轉而瞄準那些能授予強大雲端 AI 服務訪問權限的憑證。
SOCRadar 的數據分析發現，在受影響的 482 家公司中，有 295 家在監測期間的活躍日誌中出現，這表明這些暴露事件較為新近，且可能正被積極利用或交易。研究強調了這些資產為何變得如此有價值：單一遭入侵的企業 AI 憑證，可提供一個權限廣泛的集中訪問點。
與標準電郵登入不同，竊取 Microsoft Copilot 或 Azure OpenAI 部署等服務的密鑰，可能導致因欺詐性運算使用而造成即時財務損失。此外，它可能洩露敏感的專有邏輯和內部提示，並為訪問其他關聯的企業系統和數據儲存庫提供跳板。
此威脅在快速且通常去中心化的 AI 採用所造成的治理缺口下蓬勃發展。由於各部門爭相部署 AI 工具以提升生產力，驅動這些服務的憑證和 API 密鑰，其管理嚴謹度往往低於核心 SaaS 應用程式。這形成了一個「影子 AI」安全盲點，其中敏感密鑰可能被不安全地儲存、以明文記錄或隨意共享，使其容易受到竊取惡意軟件終端擷取技術的攻擊。
犯罪地下市場已注意到這一點。SOCRadar 指出，受歡迎 AI 平台的遭竊憑證，現在在非法論壇上能索要高價，反映出市場對這類新型訪問憑證的強勁需求。
對於組織，尤其是那些利用雲端 AI 獲取競爭優勢的組織而言，這項發現代表了一個關鍵的行動號召。保護 AI 資產必須納入標準安全協議。關鍵措施包括：
- 審計與盤點： 發現所有經批准及未經批准的 AI 服務訂閱、API 密鑰和服務帳戶。
- 集中化治理： 以與資料庫密碼相同的嚴謹度管理 AI 憑證，使用專用的機密管理解決方案。
- 強制執行條件存取： 應用零信任原則，包括強制性多因素認證及對受管設備的限制。
- 監控異常情況： 監察 AI 平台上不尋常的使用模式，例如大量資料查詢或非典型的登入位置。
- 輪替憑證： 實施定期輪替 API 密鑰和服務帳戶機密的政策。
竊取惡意軟件與企業 AI 憑證的融合，標誌著威脅格局的關鍵演進。隨著 AI 嵌入到業務關鍵工作流程中，保護其訪問路徑已不再可選；這是一項基本的操作必要。