Stopping Shadow AI Without Slowing Employees With Liminal

Can an organization control employee AI use without making the approved tools so restrictive that people simply work around them?

In this episode of Tech Talks Daily, I speak with Steven Walchek, Co-Founder and CEO of Liminal, about shadow AI, enterprise governance, data privacy, and the growing tension between employee productivity and corporate security.

Steven’s career includes leadership roles at FIS and AWS, along with involvement in three successful exits. His experience has given him a close view of how major technology adoption cycles begin inside companies, often before leadership has developed the policies, budgets, and controls needed to manage them.

The story behind Liminal began with an intensive period of customer discovery. Steven and his co-founder spoke with 100 prospective customers in 90 days. Across large and small businesses, they repeatedly heard concerns about what would happen to company data after employees submitted it to generative AI providers.

That concern has grown as AI tools have spread through the workplace. Steven describes two common responses from CIOs. Some permit employees to use almost any AI product, despite limited visibility into licensing terms, data retention, model training, or regulatory exposure. Others attempt to prohibit AI use completely and assume a written policy will stop employees from accessing these services.

Neither response accounts for how people behave when they believe a tool can help them work faster. An employee may use a personal account, take a photograph of a screen, or transfer information onto another device. The company has technically established a policy, but its security team may now have even less visibility into what is happening.

Steven compares this with the early adoption of cloud computing. Developers and business teams could purchase services with a credit card, while finance leaders later discovered rapidly growing AWS bills. Cloud adoption created shadow IT because people had access to useful technology before company controls caught up. Generative AI is producing a similar pattern at greater speed.

We discuss why aggressive security warnings can also produce unintended behavior. If an approved platform repeatedly frightens or reprimands employees for submitting information, they may move to an unapproved tool that creates less friction. From the employee’s perspective, the objective is usually straightforward: complete the work and produce a good result.
Steven argues that companies need an approach that gives employees a familiar AI experience while providing security teams with governance, model administration, data protection, observability, and an audit trail. He explains how Liminal attempts to combine access to several AI models with controls operating behind the user experience.

We also discuss why listening to employees matters after deployment. Steven shares how customer feedback led Liminal’s development team to change a spreadsheet feature within 24 hours. For him, that responsiveness helps businesses introduce governance without forcing people to choose between the approved system and the tool they believe can do the job.

Should enterprise AI governance begin with restrictions, or with a better understanding of what employees are trying to accomplish? Listen to the conversation and share your experience.