HiddenLayer nabs $100M as enterprises rush to secure their AI deployments

When AI information startup HiddenLayer raised its $50 cardinal Series A 3 years ago, 1 of the large questions successful the abstraction was whether the AI threats the startup was protecting against would manifest successful capable quantity to marque for a existent market. As my erstwhile workfellow Kyle Wiggers noted astatine the time, it was hard to pin down existent examples of attacks against AI astatine scale.

How rapidly things person changed. Security companies are present scrambling to physique products that tin show not just agents but also the tools and add-ons they use. While determination inactive aren’t galore headlines astir agents being exploited, the hazard of agents going haywire during accumulation is nevertheless real. And truthful the marketplace for tools to forestall that from happening has exploded: Gartner estimates companies volition walk $2.83 cardinal this twelvemonth connected products meant to unafraid AI tools, 83% much than 2025, and expects spending to scope astir $4.78 cardinal adjacent year.

HiddenLayer, which makes tools to support AI models, agents and workflows from adversarial attacks, vulnerabilities and malicious codification injections, has been capable to instrumentality bully vantage of that shift. Its co-founder and CEO, Chris Sestito, tells TechCrunch that the startup’s yearly recurring gross grew much than 10x implicit the past year. He declined to springiness an nonstop number, but said ARR is present successful the “tens of millions” of dollars, and implicit 90% of that maturation was driven by caller customers signing successful the past year.

Financial services and ample tech companies gathering AI products are presently the company’s largest verticals, and it besides has contracts with the Department of Defense and quality community. One of its customers is seemingly a “leading frontier exemplary provider” with “more than 700 cardinal play users,” which sounds similar OpenAI oregon Anthropic to me.

To marque the astir of that momentum, the startup has present raised $100 cardinal successful a Series B backing circular that was led by Delta-v Capital, with information from Ten Eleven Ventures, Morgan Stanley, Microsoft’s M12, Booz Allen Hamilton, and others.

The Austin-based startup broadly inactive sells what it utilized to successful 2023, but Sestito told TechCrunch that the biggest alteration it’s had to marque was to widen its existing products — discovery, runtime protection, onslaught simulation, and proviso concatenation information — to code punctual injection, cause manipulation and malicious instrumentality use.

“Inference is inactive inference. So whether it’s connected a accepted instrumentality learning model, whether it’s Gen AI, whether it’s an agentic enactment stream, a batch of our exertion inactive applied. So really, we haven’t had to pivot, but we’ve had to turn our scope … from accepted modeling to Gen AI to agentic,” helium said.

Sestito stressed that runtime information has particularly go a precedence arsenic AI deployments turn communal crossed businesses, and likened it to accepted endpoint detection and effect (EDR) solutions, but specifically for AI.

The company’s caller products bespeak that shift, and Sestito highlighted a caller accidental for attackers successful exploiting open-source models. “We parse and scan astir 50 antithetic AI record frameworks to marque definite that, particularly successful the lawsuit of open-source, open-weight models, that the instrumentality you’re moving with is the 1 you judge it to be, and it’s the 1 it’s purporting to be. We’re looking astatine things similar models purporting to beryllium 1 thing, but they’re different — hidden models wrong of models,” helium said.

The startup says the caller $100 cardinal is going to assistance it physique successful that direction, though with much of a absorption connected income and distribution, portion continuing to grow its engineering and research. It besides plans to grow into Europe and EMEA.

HiddenLayer whitethorn find itself tapping that warchest for more, though. Large cybersecurity companies, similar Cisco, Palo Alto Networks and Check Point often similar to bargain alternatively than physique this benignant of tech, portion startups similar Noma and Zenity person raised implicit $100 cardinal each to instrumentality connected adjacent oregon overlapping areas of this space.

Sestito acknowledged that immoderate parts of the AI information products his institution offers could yet beryllium bundled into the platforms built by companies specified arsenic Microsoft, OpenAI and AWS. However, helium expects AI infrastructure volition turn towards governance features specified arsenic discovery, individuality and argumentation controls, alternatively than the benignant of tools that his institution builds.

For now, helium says HiddenLayer’s occupation is to “scale vertically alongside artificial intelligence,” and eventually, grow horizontally into parts of cybersecurity that progressively beryllium connected AI. That’s an ambitious goal, but the startup archetypal has to beryllium it tin crook its caput commencement into an enduring concern earlier the remainder of the manufacture catches up.