DeepSeek Enterprise AI Adoption Raises New Security Questions

DeepSeek enterprise AI use appears to be growing again among U.S. companies, and the security questions are bigger than the price savings. According to 9to5Mac, new Ramp data shows DeepSeek ranked first among SaaS vendors for breakout growth relative to size across Ramp customers in May.

The detail that matters is adoption behavior. Ramp previously saw DeepSeek climb to 0.3 percent business adoption in January before falling back to 0.1 percent. The new report suggests interest has returned as companies look for cheaper alternatives to OpenAI and Anthropic, even though Ramp did not identify which companies are using the service.

This is not the same as firms running open-weight DeepSeek models on private infrastructure. 9to5Mac notes that Ramp’s data points to companies paying DeepSeek directly, meaning prompts and uploaded material may be routed through the hosted service rather than staying inside a company’s own environment.

DeepSeek enterprise AI creates data-routing risk

That distinction matters for internal documents, source code, customer records, contracts, and financial data. Anything entered into a hosted AI model is sent to the provider unless the company has strict controls and contractual protections in place.

DeepSeek’s own terms say personal data is collected, processed, and stored in the People’s Republic of China. For security teams, that raises an obvious governance issue: a low per-token price can become expensive if employees use the tool for sensitive business information.

The report also highlights a broader problem for IT departments. Employees may adopt lower-cost AI tools faster than procurement, legal, and security teams can evaluate them. That creates a shadow-AI risk similar to old shadow-IT problems, but with more sensitive data types involved.

DeepSeek may remain attractive for cost-conscious teams, especially smaller companies. But businesses using it need clear policy, data classification, and technical controls before allowing employees to paste anything sensitive into a hosted model.

You can follow more developments in Technowatt’s Artificial Intelligence coverage.

Latest News

Related News