When HiddenLayer secured its $50 million Series A three years ago, a major open question was whether the AI threats it aimed to counter would materialize at scale enough to create a viable market. At the time, concrete examples of large-scale attacks on AI systems were hard to find, as my former colleague Kyle Wiggers pointed out. How dramatically the landscape has shifted since then. Today, security firms are rushing to develop products capable of monitoring not just AI agents but also the tools and plugins they rely on. Although headlines about exploited agents remain scarce, the risk of agents malfunctioning during real-world production is nonetheless genuine. Consequently, the market for protective solutions has surged: Gartner projects companies will invest $2.83 billion this year in AI security products, an 83% jump from 2025, with spending expected to climb toward $4.78 billion the following year.

HiddenLayer, which builds tools to shield AI models, agents, and workflows from adversarial attacks, vulnerabilities, and malicious code injections, has capitalized on this momentum. While the founder declined to disclose exact figures, he noted that annual recurring revenue now sits in the "tens of millions" of dollars, with over 90% of that growth stemming from new customers signed in the past year. Financial services and major tech firms developing AI products represent the company’s largest verticals, and it also holds contracts with the Department of Defense and the intelligence community. One notable client appears to be a "leading frontier model provider" boasting "more than 700 million weekly users," which strongly suggests either OpenAI or Anthropic.

To leverage this upward trajectory, HiddenLayer has raised $100 million in a Series B round led by Delta-v Capital, with participation from Ten Eleven Ventures, Morgan Stanley, Microsoft’s M12, Booz Allen Hamilton, and others. "Inference remains inference," the CEO explained. "Whether it’s a traditional machine learning model, generative AI, or an agentic workflow, much of our technology still applies. So we haven’t had to pivot—we’ve had to expand our scope, moving from conventional modeling to Gen AI and then to agentic systems." He emphasized that runtime security has become a top priority as AI adoption spreads across industries, likening it to endpoint detection and response (EDR) solutions, but tailored specifically for AI environments.

The company’s newest offerings reflect this evolution, with a particular focus on a fresh attack vector: exploiting open-source models. "We parse and scan around 50 different AI file frameworks to ensure—especially with open-source, open-weight models—that the tool you’re using is genuinely what it claims to be," he said. "We’re looking for models that pretend to be one thing but are actually another, hidden models nested inside models."

HiddenLayer says the new $100 million infusion will drive growth in that direction, with an emphasis on sales and distribution while continuing to bolster engineering and research. The startup also plans to expand into Europe and the EMEA region. However, it may need to dip into that war chest further. Large cybersecurity players like Cisco, Palo Alto Networks, and Check Point often prefer acquiring such technology over building it internally, while startups such as Noma and Zenity have each raised over $100 million to tackle adjacent or overlapping challenges in this space.

The CEO acknowledged that some of HiddenLayer’s offerings could eventually be integrated into platforms from giants like Microsoft, OpenAI, and AWS. Yet he anticipates that AI infrastructure will lean more toward governance features—such as discovery, identity, and policy controls—rather than the security tools his company develops. For now, HiddenLayer’s mission is to "scale vertically alongside artificial intelligence," and later broaden horizontally into cybersecurity areas increasingly reliant on AI. It’s an ambitious vision, but the startup must first demonstrate it can convert its early advantage into a lasting business before the wider industry catches up.