cyber.hendryadrian.com
Interesting Stuff · jamf.com

AI Governance Must Address Data Exposure, Retrieval Risks and Prompt Injection

· Original article ↗

Summary

An explainer outlines enterprise AI security risks, including shadow AI, stale indexed data and indirect prompt injection, and recommends operational controls for access, data handling and oversight.

Key points

  • The article argues that AI tools can expand enterprise data exposure and attack surfaces when governance is limited to policies or compliance reviews.
  • It cites reports that Samsung restricted employee use of ChatGPT after engineers reportedly pasted sensitive internal information into the public tool; this was not described as an external hack.
  • It recommends monitoring AI use, applying DLP to prompts and uploads, classifying input data, and distinguishing approved enterprise accounts from personal accounts.
  • The article says more than 20,000 repositories that were once public reportedly remained accessible through Copilot via Bing caching, affecting more than 16,000 organizations.
  • It recommends enforcing current user permissions during AI retrieval and ensuring indexes respect deletion and permission changes, with secret scanning and auditing.
  • It describes indirect prompt injection, including EchoLeak (CVE-2025-32711), as a risk in which malicious content may influence an AI assistant to disclose information.
  • Suggested safeguards include treating external content as untrusted, isolating prompt scopes, limiting tool use, filtering outputs and continuously testing systems.

Article Details

Topic
Enterprise AI governance and security risks, including shadow-AI data exposure, retrieval of previously public data, and indirect prompt injection.

CVE

CVE-2025-32711EchoLeak, tracked as CVE-2025-32711, is a major example. NVD classifies the issue as AI command injection in Microsoft 365 Copilot that could allow unauthorized information disclosure over a network. Research on

Vendors

MicrosoftA notable case involved Microsoft Copilot and GitHub repositories. According to TechCrunch, more than 20,000 repositories that had once been public, but were later deleted or made private, still had data accessible

Products

ChatGPTOne of the earliest warnings came from Samsung’s reported ChatGPT data-leak incident in 2023. The company reportedly restricted employee use of ChatGPT and similar tools after reports that engineers pasted sensitiveGitHubA notable case involved Microsoft Copilot and GitHub repositories. According to TechCrunch, more than 20,000 repositories that had once been public, but were later deleted or made private, still had data accessibleMicrosoft 365 Copilotas CVE-2025-32711, is a major example. NVD classifies the issue as AI command injection in Microsoft 365 Copilot that could allow unauthorized information disclosure over a network. Research on EchoLeak

Related Articles