HiddenLayer raises $100M to secure AI agents at runtime

Why it matters

HiddenLayer's round is a wager that enterprises will buy an independent security layer for AI agents instead of relying on model providers and incumbent cybersecurity suites to bundle one.

Chris Sestito, Tanner Burns and Jim Ballard's HiddenLayer raised a $100 million Series B to expand security software built for AI models, autonomous agents and the tools those agents can access, TechCrunch reported on September 2nd.

Delta-v Capital led the financing. Ten Eleven Ventures, Morgan Stanley, Microsoft's M12 and Booz Allen Ventures participated, according to HiddenLayer's announcement. The round brings HiddenLayer's publicly announced funding to at least $156 million, following a $6 million seed round in 2022 and a $50 million Series A in 2023.

HiddenLayer said it will put the new capital into product development, engineering, research, sales and distribution, with international expansion beginning in Europe and the wider EMEA region. Its immediate product focus includes runtime monitoring for AI agents and Agent Harness Security, software that watches coding agents as they read files, execute commands, install dependencies and interact with repositories.

A founding attack became the product thesis

HiddenLayer grew from an attack Sestito, Burns and Ballard encountered while working at Cylance, the AI-based antivirus provider BlackBerry acquired in 2019. In a 2024 submission to the National Institute of Standards and Technology, HiddenLayer said the three founders responded to an inference attack that exposed weaknesses in Cylance's machine-learning model and allowed attackers to evade detection.

That experience gave the founders an early view of a problem enterprise buyers had barely begun budgeting for: machine-learning systems introduced their own attack surface, separate from the servers and applications around them. HiddenLayer launched in 2022 with products for finding AI assets, scanning model files, simulating attacks and detecting malicious behavior while models were running.

The underlying bet has stayed intact while the target has expanded. Sestito told TechCrunch that HiddenLayer extended its products from traditional machine-learning models into generative AI and autonomous agents, adding defenses against prompt injection, agent manipulation and malicious tool use.

Open-weight models create another distribution channel for hostile code. Sestito said HiddenLayer scans roughly 50 AI file frameworks for tampered artifacts, including model files that conceal other models or do not contain what their labels claim. That resembles software supply-chain scanning, applied to files that enterprises increasingly download from public model repositories and insert into production systems.

HiddenLayer reports a 10x revenue year

HiddenLayer told TechCrunch that annual recurring revenue grew more than tenfold over the past year and reached the "tens of millions" of dollars. Sestito did not provide a precise ARR figure. He said more than 90% of the growth came from customers added during the year.

HiddenLayer separately said it signed more than 50 new platform customers across banking, insurance, pharmaceuticals, technology, government and other sectors. HiddenLayer also reported contracts with the US defense and intelligence communities and said an unnamed frontier-model provider uses its products to protect services reaching more than 700 million weekly users. HiddenLayer did not identify that customer, so the claim cannot establish which model provider bought the software or the scope of the deployment.

The revenue figures explain the size of the round, while the spending plan shows where the pressure sits. HiddenLayer already built much of the technical platform. The next contest is distribution: getting embedded in enterprise security budgets before large cybersecurity vendors bundle comparable controls into existing platforms.

The category is attracting larger checks and acquirers

HiddenLayer is entering that fight with heavily financed competitors. Noma Security raised a $100 million Series B in July 2025 and later said its total backing exceeded $130 million. Zenity raised a $125 million Series C on August 3rd, 2026, to build security products for autonomous AI systems.

Cybersecurity incumbents have been buying their way into adjacent markets. Palo Alto Networks completed its acquisition of Protect AI in July 2025. Cisco built its Foundation AI group from Robust Intelligence and completed its acquisition of agent-identity specialist Astrix Security in June 2026.

HiddenLayer therefore has to prove that AI-specific runtime protection remains a durable standalone category. Sestito acknowledged to TechCrunch that Microsoft, OpenAI, AWS and other infrastructure providers could eventually bundle pieces of AI security into their platforms. The $100 million round gives HiddenLayer more time to establish its software as the independent control layer before those platforms, or the cybersecurity vendors acquiring around them, close the gap.