AI assistants are a little bit of a double edged sword in business. Employees can feed the LLM data and ask it to perform basic tasks, and it will get the job done. However, sometimes people are a little too eager to use AI and end up feeding the LLM data that really shouldn't be shared outside of the business.

As such, companies are developing apps that act as a middle man between the user and the AI that double-checks for any sensitive data. One such company is Tailscale, which has developed a new app called Aperture to keep an eye on what people are feeding into an AI.

Aperture keeps an eye on what data employees are feeding into AI

People are a little too eager to share information with AI

In a press release, Tailscale revealed it's new Aperture tool. It sits between the user and the LLM and double checks everything the user is sending. If the tool detects the user is sending over data they're not meant to, Aperture will step in and prevent the message from going through. It also reports on who sent the data so that employers can keep an eye out for who might be leaking data.

So, how big a problem is this? Well, Tailscale claims that it's actually pretty dire:

Recent analysis of workplace AI usage found 34.8% of corporate data employees put into AI tools is sensitive, and that source code is the most common category of sensitive data going into AI. Separately, a global study led by the University of Melbourne with KPMG found 48% of all workers reported uploading sensitive company data into public AI tools. Cisco Talos reported 1,100+ publicly exposed Ollama LLM servers, noting it took minutes to identify most of them.

Aperture also makes it easier to keep tabs on what's going on. It ties prompts and commands to specific users, so if someone is misusing the LLM, the company knows exactly who did it. And Aperture can hold onto LLM API keys, meaning people can use the models without needing to pass around a key and reducing the chances of it leaking out to unwanted users.