Tools for spotting and disabling AI systems in an enterprise
Source Kaspersky Security Blog
Description
Key methods for cutting off AI access to an organization’s core IT assets.
Impact
The impact involves potential data leakage from corporate systems to external AI service providers due to unsanctioned AI tool usage. This can expose sensitive corporate information outside the organization’s control. The risk varies by AI tool type and corporate policy enforcement. There are no known exploits in the wild and no direct software vulnerability. The threat is primarily operational and compliance-related, affecting data confidentiality and corporate governance.
Defensive guidance
No software patch is applicable. Mitigation focuses on policy enforcement and technical controls: 1) Identify existing AI tools in the environment using DNS, web gateways, endpoint protection, application control, browser control, and SaaS security posture management tools. 2) Disable or restrict AI tools using built-in corporate policy settings where available. 3) Implement hard blocks at endpoint or network levels to prevent unauthorized AI tool use. 4) Manage OAuth consent settings in platforms such as Microsoft 365, Google Workspace, Salesforce, and Slack to prevent users from granting unauthorized AI apps access to corporate data. 5) Regularly audit approved AI applications and adjust policies based on organizational risk tolerance and regulatory requirements. Follow vendor guidance as detailed in the Kaspersky blog post.
Summary
The threat concerns unauthorized or unsanctioned AI tools operating within enterprise IT environments, which can cause data leakage by sending corporate data to external AI service providers. AI tools are categorized into four types: platform-native AI features embedded in software, AI companions integrated into business applications, standalone AI chatbots often installed without approval, and desktop-native AI agents with broad access. Detection methods include DNS filtering, web gateways, endpoint protection platforms, application and browser controls, and SaaS security posture management focusing on OAuth permissions. The main mitigation approach is to enforce corporate policies that disable or restrict AI tool usage, block AI traffic at network or endpoint levels, and tightly control OAuth consent to prevent unauthorized data access. The vendor (Kaspersky) provides detailed guidance but no software patch is applicable since this is a policy and configuration challenge rather than a software vulnerability.