AI is now the number one channel for data exfiltration in the enterprise

Artificial intelligence has become the single largest uncontrolled channel for corporate data exfiltration, surpassing both shadow SaaS and unmanaged file sharing, according to a new report from AI and browser security company LayerX. The research, based on real-world enterprise browsing telemetry, indicates that the primary risk from AI in the enterprise is not a future threat, but a present-day reality unfolding in everyday workflows.

Sensitive corporate data is already flowing into generative AI tools like ChatGPT, Claude, and Copilot at high rates, primarily through unmanaged personal accounts and the copy-and-paste function.

The rapid, ungoverned adoption of AI

AI tools have achieved a level of adoption in just two years that took other technologies decades to reach. Nearly half of all enterprise employees (45%) already use generative AI, with ChatGPT alone reaching 43% penetration. AI now accounts for 11% of all enterprise application activity, rivaling file-sharing and office productivity apps.

This growth has largely occurred without corresponding governance. The report found that 67% of AI usage happens through unmanaged personal accounts, leaving security teams with no visibility into which employees are using which tools or what data is being shared.

Sensitive data is leaking through files and copy-paste

The research uncovered alarming trends in how sensitive data is being handled with AI platforms.

• File uploads: 40% of files uploaded into generative AI tools contain personally identifiable information (PII) or payment card industry (PCI) data. Nearly four in ten of these uploads are done using personal accounts.
• Copy-and-paste: The primary channel for data leakage is the copy-paste function. 77% of employees paste data into generative AI tools, and 82% of that activity comes from unmanaged personal accounts. On average, employees paste sensitive data into these tools via personal accounts at least three times per day.

The report identifies copy-and-paste into generative AI as the number one vector for corporate data leaving enterprise control. Traditional security programs focused on scanning file attachments and blocking unauthorized uploads are missing this threat entirely.

Other major security blind spots

The report highlights two other critical areas where corporate data is at risk.

• Non-federated logins: Even when employees use corporate credentials for high-risk platforms like CRM and ERP systems, they overwhelmingly bypass single sign-on (SSO). 71% of CRM logins and 83% of ERP logins are non-federated, making a corporate login functionally the same as a personal one from a security visibility standpoint.
• Instant messaging: 87% of enterprise chat usage occurs through unmanaged personal accounts, and 62% of users paste PII or PCI data into them.

Recommendations for enterprise security in the AI era

The report offers several clear recommendations for security leaders.

• Treat AI security as a core enterprise category, not an emerging one, with monitoring for uploads, prompts, and copy-paste flows.
• Shift from a file-centric security model to an action-centric one that accounts for file-less methods like copy-paste and chat.
• Restrict the use of unmanaged personal accounts and enforce federated logins for all corporate applications.
• Prioritize the highest-risk application categories for the tightest controls: AI, chat, and file storage.

The findings paint a clear picture: the enterprise security perimeter has shifted to the browser, where employees fluidly move sensitive data between sanctioned and unsanctioned tools. The report concludes that if security teams do not adapt to this new reality, AI will not just shape the future of work, but also the future of data breaches.

Featured image credit