New Plug-In Helps Companies Prevent Sensitive Data Leakage to AI Chatbots
A plug-in, called LLM Shield, has been released by San Francisco-based start-up Patented.ai to help companies protect their sensitive or proprietary data from AI chatbots. The LLM Shield is an AI model designed to recognize various kinds of sensitive information including trade secrets, personally identifiable information, HIPPA-protected health data, and military secrets. When an employee enters such sensitive data into a chatbot, LLM Shield is able to recognize it and alert the employee that they are about to share sensitive information in the chatbot. The plug-in will not charge the customer for such warnings, but will charge if it helps the customer block a piece of content from being shared or obfuscates certain aspects of sensitive content.
According to Patented.ai, while AI is a powerful tool, it’s not necessary to ban employee access. Rather, with better visibility and control, companies should have confidence in their employees and their usage of LLMs. Based on the demand prompted by leaks at companies like Samsung, this development of LLM Shield was accelerated. Patented.ai intends to roll out new types of security products powered by its AI model throughout this year, which will include a version of the plug-in for personal use.
Generative AI models have many applications within businesses large and small. However, many companies are becoming increasingly concerned about how sensitive or private information could be used by the models. The inner workings of such AI models are still not completely understood. In an attempt to minimize the chance of leaks, some major financial institutions have banned the use of OpenAI‘s ChatGPT, including Bank of America, Goldman Sachs, Citigroup Inc., Deutsche Bank AG, and Wells Fargo. Avivah Litan, an analyst with consulting firm Gartner, believes a new class of AI trust-, risk-, and security-management tools is needed to manage data flows and processes between users and companies who host generative-AI foundation models.
