Generative artificial intelligence tools like ChatGPT became all the rage in 2023, with businesses and educational institutions determining how to implement or avoid using these tools. Several regulated industries and companies, including Apple and Samsung, have even banned their employees from using generative AI tools, fearing that sensitive internal data or proprietary information will be leaked.
A Denver startup aims to address these security concerns.
Liminal, which launched in late January, is a business-to-business AI security startup for regulated industries — such as health care, financial services, education, insurance, defense and life sciences. Liminal’s model-agnostic and horizontal approach lets customers securely use AI tools like ChatGPT and Azure OpenAI on one platform.
“In the regulated industry, security is the number one blocker to any real and scaled deployment of generative AI,” Liminial CEO and co-founder Steven Walchek told Colorado Inno. “... We wanted to become an enabler in generative AI but we wanted to do it in a secure and safe manner.”
Walchek, a serial entrepreneur who founded his first company when he was 19, began beta testing Liminal in the fall of 2023. Today, its platform is used by one of the largest pharmaceutical companies in the world and one of the largest health systems in the U.S. Liminal is also currently used by large and mid-sized banks, according to the startup.
Walchek said he has founded or co-founded three companies. He’s also served in three other leadership positions, gone through two acquisitions and helped create FIS Impact Labs, an in-house incubator for future fintech solutions. When the AI revolution began, he knew he wanted to build something new in the AI space.
After speaking with several chief information officers and chief information security officers at large regulated companies, Walchek found a problem he wanted to fix.
“What emerged was this problem around security,” he said. “And they broke down that security problem for me into three categories, which first was regulatory compliance risk ... the second was the notion of corporate data security that IT didn’t want to see that leak in any way shape or form to a large language model, and then the third was just reputational.”
Addressing regulatory compliance risk and security are two things Liminal set out to address.
Here’s how it works. Liminal offers three different secure interfaces. One allows users to communicate with a generative AI tool, similar to how people use ChatGPT. However, Liminal identifies sensitive information such as a person’s name, date of birth, patient-specific medical terminology and email addresses, and desensitizes it.
“The algorithm that we have built to detect sensitive information is very, very robust,” said Aaron Bach, Liminal’s co-founder and chief technology officer. “It’s able to detect these concepts in context.”
Liminal then removes this sensitive information and replaces it with nonsensitive information when submitting an inquiry to a generative AI tool. When a response is generated, Liminal can re-add that sensitive information back to the end response.
It works with any public or private generative AI model that a customer or business has licensed.
“The notion is our customers are using multiple models for multiple purposes,” Bach said “So, we allow them one spot to configure all the models that they’re going to use, and once you configure a model here, it becomes available throughout all those experiences.”
Liminal works with Microsoft Copilot, too.
If they are customers, developers also have access to Liminal’s code, giving them greater control over each tool.
Liminal doesn’t manage authentication or authorization. Instead, it integrates with a company’s existing user management and identification system. Bach said this allows businesses to control who accesses the generative AI tools on Liminal’s platform.
When a business becomes a Liminal client, they get access to the startup’s entire platform.
“Generative AI will continue to proliferate and become increasingly specialized,” Eric Tobias, a partner with High Alpha, said in a statement. “We have strong conviction in Liminal’s approach to solving for the critical barrier to organizational adoption. ... Their successes to date are empowering a new era of value creation driven by unlocking generative AI for enterprises in regulated industries.”
Liminal has raised $3.15 million in safes, or simple agreements for future equity, from High Alpha and Matchstick Ventures.
The startup’s team consists of 14 employees, about half of whom are based in Denver. Walchek said he is actively hiring three more people, adding that he didn’t set out to build a team in Denver but found good talent here.
Liminal's CEO Steven Walchek has founded or co-founded three companies.
