Employee-Created personalized AI applications pose security dangers
The latest findings from Netskope indicate a significant 50% increase in the usage of Generative AI (GenAI) platforms among enterprise end-users. This surge is primarily driven by the growing demand from employees for tools that facilitate the development of custom AI applications and agents. Despite efforts to safely enable Software as a Service (SaaS) GenAI applications and AI agents, the rise of shadow AI—unsanctioned AI applications used by employees—continues to pose substantial security risks. It is estimated that over 50% of all current app adoption falls under the category of shadow AI.
GenAI platforms serve as foundational infrastructure tools that empower organisations to create custom AI applications and agents. They represent the fastest-growing segment of shadow AI due to their user-friendly nature and flexibility. These platforms enable a direct connection between enterprise data stores and AI applications, which raises new data security concerns. Consequently, there is an increased emphasis on Data Loss Prevention (DLP) and continuous monitoring. Network traffic associated with GenAI platform usage has surged by 73% over the past three months. As of May, 41% of organisations reported using at least one GenAI platform, with approximately 29% utilising Microsoft Azure OpenAI, 22% using Amazon Bedrock, and 7.2% employing Google Vertex AI.
Ray Canzanese, Director of Netskope Threat Labs, emphasises that the rapid growth of shadow AI necessitates organisations to identify who is creating new AI applications and agents, as well as where these are being developed and deployed. Security teams aim to support employee innovation without hindering it, but the increasing use of AI requires organisations to enhance their AI application controls and adapt their DLP policies to include real-time user coaching elements. Furthermore, organisations are exploring various options for deploying GenAI locally through on-premises GPU sources and developing tools that interact with SaaS GenAI applications.
Currently, 34% of organisations are using Large Language Model (LLM) interfaces, with Ollama leading in adoption, while others like LM Studio and Ramalama are just beginning to gain traction. Employee end-users are actively experimenting with AI tools and frequently visit AI marketplaces. For instance, 67% of organisations report that users are downloading resources from Hugging Face. The potential of AI agents is driving this trend, as data reveals a critical mass of users across organisations are now building AI agents and leveraging the agentic AI features of SaaS solutions. GitHub Copilot is currently utilised by 39% of organisations, with 5.5% of users running agents generated from it.
