PromptArmor Is Building the Security Layer for Third-Party AI

Most companies don’t build the AI they use.

It arrives through the software they already buy. A SaaS vendor adds an AI assistant. A new application connects a model to sensitive company data. An employee wants to use an AI tool that security hasn’t reviewed. Underneath those products can sit models, connectors, infrastructure providers, and data processors that are changing constantly.

For enterprises, that creates a new version of an old problem: how do you know whether the technology you’re trusting is actually safe?

PromptArmor is building the security infrastructure to answer that question.

AI Changes the Vendor Review

PromptArmor started from the technology itself.

Before launching the company, Shankar Krishnan and Vikram Jayanthi pentested more than 50 LLM applications. They kept finding many of the same vulnerabilities: opportunities for data exfiltration, phishing, system manipulation, and other attacks that traditional application security wasn’t designed to catch.

That work became the foundation for PromptArmor. The company initially focused on protecting LLM applications directly and continued researching how widely deployed AI products could be exploited. Its research has since uncovered vulnerabilities across products including Slack AI, Writer, Snowflake Cortex, Claude, and other major AI platforms.

But the problem has grown considerably as AI has moved into the enterprise.

Companies are no longer evaluating a handful of standalone AI tools. AI is appearing inside the software they already use, and those products are changing quickly. Vendors introduce new models, turn on new features, expand data access, and add connectors that give AI systems the ability to reach into other parts of the organization.

That makes the traditional vendor review increasingly incomplete.

PromptArmor gives security and third-party risk teams a way to understand the AI inside the products their companies use, assess the risks those systems introduce, and monitor how those risks change over time. The company evaluates vendors across AI-specific security, privacy, and legal risks, while continuously watching for changes that could alter a product’s risk profile.

The goal is straightforward: give enterprises enough visibility to adopt AI without flying blind.

The Risk Beneath the Application

Very few AI products exist in isolation.

An enterprise might buy software from one company that relies on a model from another, connects to Slack or Salesforce, accesses internal company data, and gives an AI agent permission to take actions across several other systems.

Each connection makes the product more useful. It can also expand what happens if something goes wrong.

PromptArmor’s research into indirect prompt injection shows why. An AI system can ingest malicious instructions hidden inside content it retrieves, potentially causing the model to expose sensitive information or take unintended actions through the tools connected to it. As AI applications gain access to more data and more systems, understanding those relationships becomes increasingly important.

The underlying technology is also unusually dynamic. PromptArmor has documented how connectors and AI capabilities can change frequently, sometimes altering what data a system can reach or what actions it can perform. A security review completed months ago may no longer describe the product running inside the company today.

PromptArmor is building a living view of that environment: which vendors are using AI, what systems sit underneath them, what information they can access, which risks they introduce, and what has changed since they were approved.

As AI spreads across the enterprise, that map becomes more valuable.

A Team That Saw the Problem Early

Shankar and Vikram met at Berkeley, where Shankar studied data science and economics and Vikram studied computer science and public policy. Their careers gave them complementary exposure to the systems PromptArmor now has to understand.

Shankar worked across companies including Tesla, Observe, Snappr, and Accenture. Vikram worked in infrastructure and security at Google, Roblox, and Symantec. By the time they started PromptArmor, they had experience across security, infrastructure, data, and the kinds of large software systems where AI was beginning to appear.

What stands out even more is how early they started digging into the security implications of language models.

Pentesting dozens of LLM applications before most enterprises had developed an AI security strategy gave them a firsthand view into how these systems fail. Their subsequent research has kept them close to the edge of the problem as models have gained access to more data, tools, and applications.

That technical depth matters in third-party risk. PromptArmor isn’t simply collecting questionnaires about whether a vendor uses AI. The company has built its own expertise around how AI systems can actually be attacked, then uses that knowledge to help enterprises evaluate the products they’re being asked to trust.

Shankar and Vikram started by breaking AI applications.

Now they’re using what they learned to help companies deploy them safely.

A New Layer of Enterprise Security

Enterprise security has repeatedly expanded around major shifts in computing.

The move to the cloud created new security categories because infrastructure left the corporate data center. SaaS created another wave as company data spread across hundreds of outside applications. AI is now introducing another set of dependencies that enterprises need to understand.

The scale of that surface area could become enormous. AI is being added to software across nearly every category while companies simultaneously adopt standalone models, applications, agents, connectors, and infrastructure. Each addition can introduce new access to company data and new ways for software to act on it.

Someone inside the enterprise has to understand what all of those systems are allowed to see and do.

PromptArmor has the opportunity to become an important source of intelligence for that decision. The company already serves security and risk teams evaluating AI across large enterprises and law firms, and its platform spans vendor assessment, continuous monitoring, and AI-specific threat intelligence.

The more AI companies use, the more important it becomes to know exactly what they’re trusting.

Why We Invested

We invested in PromptArmor because Shankar and Vikram saw the security implications of language models early and went deep enough technically to understand where the real vulnerabilities were.

They also chose a problem that becomes more important as AI succeeds. Every new model, agent, connector, and AI feature gives companies more capability, but it also introduces another system that may touch sensitive data or take consequential actions.

Enterprises aren’t going to stop adopting those products. They need security infrastructure capable of keeping up with them.

Shankar and Vikram have been studying how AI systems fail since before most companies knew which questions to ask. PromptArmor turns that expertise into something enterprises can use every day.

The company started by breaking AI applications.

Now it’s building the intelligence layer that helps enterprises decide which ones they can trust.

Visit PromptArmor at https://promptarmor.com.