HiddenLayer $100M Series B Funding Round

The investor list on this round is the real story. Delta-v Capital leads, but the participants - Morgan Stanley (Wall Street strategic capital), Microsoft's M12 (Big Tech strategic capital, back for a second round), and Booz Allen Hamilton (a firm whose core business is federal and defense contracting) - rarely converge this fast around one narrow security niche ^[1]. Gartner now estimates enterprises will spend $2.83 billion on AI security products this year, 83% more than in 2025, climbing toward $4.78 billion next year ^[1]. That trajectory is what a bank's strategic investment arm needs to see before writing a check into a category it wouldn't have touched three years ago. Morgan Stanley's Zheng Wang put it plainly: HiddenLayer was early to recognize that enterprise security is critical to the safe deployment of AI. The composition of the cap table, not just its size, is the signal that AI security has moved from novelty spend to a defended, budgeted line item.

HiddenLayer's original pitch was about protecting AI models from theft, tampering, and adversarial manipulation. The Series B money is aimed squarely at a newer, narrower problem: the AI coding agent itself. Roughly a month before this raise, the company unveiled Agent Harness Security, extending its Runtime Security module to guard against prompt injection, secret exposure, and unsafe command execution inside AI coding agents as they run ^[4]. That product line, alongside the broader Agentic Runtime Security push, is explicitly what the new capital is meant to scale ^[2], alongside sales and distribution and a move into Europe and EMEA ^[1]. It is a bet that the next wave of enterprise risk will not come from a model file sitting in a repository, but from an autonomous coding agent that has shell access, credentials, and the ability to execute commands on its own - a much harder thing to sandbox than a static model artifact.

When HiddenLayer closed its $50 million Series A in September 2023 ^[3], the open question hanging over the company was whether AI-specific threats - model theft, data poisoning, adversarial prompt attacks - would actually materialize at enterprise scale, or stay a theoretical concern for security teams to file away. Three years later that uncertainty has resolved: annual recurring revenue grew more than 10x in the past year alone, reaching the tens of millions of dollars, with over 90% of that growth coming from entirely new customers rather than expansion of existing accounts ^[1]. The company's new customer base spans sectors including the US defense and intelligence communities, and it counts a frontier model provider serving more than 700 million weekly users among its customers ^[2]. Landing that kind of reference customer while a market is still being defined is unusual, and it is a large part of why the round doubled the size of the prior one.

The technical thesis underpinning HiddenLayer's expansion is narrower than it sounds. CEO Chris Sestito's explanation for why a company built to scan traditional machine learning models can pivot into securing generative AI and now autonomous coding agents comes down to a single idea: inference is still inference, whether it happens on a classical ML model, a GenAI system, or an agentic workflow, so a large share of the underlying detection technology still applies ^[1]. That is a bet that the attack surface - the moment a model or agent processes an input and produces an action - stays structurally similar even as the systems built on top of it get more autonomous. It is also the argument the company needs to make to justify spending $100 million on breadth (agentic runtime, coding-agent harnesses, international expansion) rather than narrowing focus. HiddenLayer has also appeared in an AWS-hosted session discussing securing agentic AI workloads and Model Context Protocol deployments.