Unbound raises $4 million funding to help enterprises embrace AI tools on their terms

As generative AI tools proliferate across the workplace, many enterprises are finding themselves unprepared for the security, governance and cost implications of their uncontrolled use.

Employees are quietly relying on copilots for code, content, and campaign ideas — often without any oversight from IT departments. That growing disconnect between adoption and control has opened a new niche in the enterprise AI infrastructure market. 

Unbound, a startup founded by former Adobe, Palo Alto Networks, and Imperva veterans, has raised $4 million in a seed funding round to close that gap. The round was led by Race Capital and included participation from Y Combinator, Wayfinder Ventures, Massive Tech Ventures, and several notable angels such as Ram Shriram, a founding board member at Google, and Eli Brown, the CEO of Guilded.

Founded by Rajaram Srinivasan and Vignesh Subbiah, Unbound has developed an AI Gateway platform that lets IT teams monitor, control, and optimize the use of generative AI tools across their organizations. The product intercepts prompts from tools like Cursor, Roo, and internal copilots, scanning for sensitive data, enforcing access policies, and directing requests to the most appropriate model — including internal, open-source alternatives.

The need for a solution like Unbound became apparent to the founders during the early days of GPT-3.5 adoption, when they observed teams leaking passwords, personally identifiable information, and other proprietary content into external AI models. Legacy data loss prevention tools proved ill-suited for this emerging workflow, either blocking access entirely or failing to recognize the risk embedded in natural language prompts.

Unbound offers a middle ground. Instead of blanket bans, its system redacts sensitive content in real time and reroutes high-risk prompts to safer models hosted within the organization’s cloud. In doing so, it enables enterprises to maintain productivity while enforcing guardrails on data privacy and compliance.

According to the startup, the platform has already prevented over 7,000 potential data leaks, including the exposure of hundreds of secret credentials and patient records. At the same time, it has helped customers reduce AI-related costs by up to 70% through smarter routing of workloads to lower-cost models.

“As AI tools become mainstream, enterprises are turning to flexibility and control,” said Srinivasan, Unbound’s CEO. “They want visibility into what’s being used, assurance that their data is protected, and the ability to swap in better models as the space evolves.”

Subbiah, who serves as CTO, said, “Defaulting to blanket bans on AI tools is like being in the times of GPT-3.5. Unbound enables surgical security controls into every AI request so teams can innovate freely without putting corporate secrets at risk.”

“AI is projected to reach $4.8 trillion in market value for the enterprise by 2033 globally — but without proper guardrails, that value is at risk. From shadow models to data leaks, the dangers of unmanaged AI are very real.  We are excited to back Rajaram Vignesh and the Unbound Security team as they create a new category of AI infrastructure: one built for safety, observability and cost discipline from day one,” said Edith Yeung, General Partner at Race Capital.

As part of its growth plans, Unbound will deploy over $1 million in India to support hiring. The startup currently has a team of eight based in Bengaluru and aims to scale operations in the country, where demand for cost-efficient and performance-focused AI solutions is growing.

"India is known for its cost efficiency in achieving great outcomes. For example, our Mars mission Mangalyan was cheaper than the movie gravity. We’re seeing similar trends with LLMs. You can’t have a system that costs $7 to troubleshoot one issue. Our customers have achieved over 70% cost savings through smart routing without compromising on performance,” Srinivasan added.