AI monitoring startup Lanai Inc. says it’s ready to tackle the growing risk presented by “shadow” artificial intelligence with the launch of its new observability agent, which runs so-called “detection models” directly on corporate devices, instead of routing conversations through centralized cloud infrastructure.

Lanai says its AI Observability Agent is the next step in AI governance, eliminating the risks associated with employees who use unauthorized chatbots and services. It provides full visibility into employee’s generative AI interactions across any application, without sending data outside of an organization’s own network.

The startup believes that “shadow AI” has become an acute crisis for enterprises, citing a study by LayerX Ltd. that shows more than 89% of all enterprise AI usage is completely invisible to information technology and security teams. That’s because employees are doing things such as using personal ChatGPT accounts for work to increase their productivity, while software engineers employ unauthorized tools such as Codeium and Cursor to help write code faster. In addition, many workers blithely feed sensitive information into the AI features embedded in platforms such as Salesforce, Adobe and Microsoft 365, without their supervisors knowing about it.

That’s problematic, according to Lanai co-founder and Chief Executive Lexi Reese, because it means security teams are being asked to secure things that they cannot see. “The shadow AI problem is exploding as employees are using personal accounts, unapproved coding agents and embedded AI features in everyday SaaS tools, and this all sits outside IT’s visibility,” she said. “Traditional tools might catch someone visiting ChatGPT.com, but they have no idea if that employee just shared company trade secrets or a client’s sensitive revenue information.”

Reese said Lanai’s AI Observability Agent embeds lightweight observability models onto every employee’s device to bring this activity out of the shadows and add guardrails to ensure that sensitive information is not being shared with third-party AI, ensuring safety without sacrificing speed or innovation.

Lanai’s observability agents can be integrated with existing data management tools within 24 hours or less, and enable dynamic detection across any application employees use, without the need to create static lists. They carry out prompt and response analysis in real time to ensure no sensitive data or workflow insights are being leaked, with all of the processing done on the employee’s device.

By using Lanai as a source of intelligence into what AI tools people are using, companies can approve the services that are most useful and shut down others to protect vast amounts of data, the startup said.

To demonstrate how effective its platform is, Lanai gave some insights into how early adopters have been able to scale back shadow AI without hurting productivity. For instance, it cites one Fortune 500 tech company, which found that its software engineers were using 12 different coding agents, feeding them volumes of sensitive information on a daily basis. That customer used Lanai’s agents to identify the best performing coding agents and work out which ones were suitable for approval.

Those that met its requirements were then scaled across departments, while unauthorized tools were blocked. The result: Productive AI usage soared by 300%.

Similarly, a healthcare organization found that its more than 20,000 employees were using 34 different AI tools, despite only five being approved. That included doctors feeding patient data into their personal ChatGPT accounts and developers using unauthorized agents to write code for regulated software applications. By implementing automated controls, it saw an 80% reduction in data exposure incidents in just 90 days.

Lanai also points to the example of an unnamed financial services firm, which discovered that 73% of its total AI usage was through personal accounts. It shut down unauthorized usage and identified the best-performing approved AI tools to help the company eliminate $700,000 in annual costs on unnecessary services.

Lanai’s other co-founder Steve Herrod said the company has ambitions to become the “Datadog of AI” by identifying and securing the AI services that bring the most benefits to enterprises, while shutting down the rest.

“We’re moving AI observability from the network to the edge,” he said. “It’s like the shift from monitoring server rooms to having telemetry inside every virtual machine. Whereas traditional approaches see network traffic, we see the actual prompts and responses, which are where the real risks and value live.”

Image: SiliconANGLE/Microsoft Designer