HiddenLayer, a US company focused solely on AI security, has raised 100 million USD (about 15.6 billion yen) in a Series B round. Delta-v Capital led the round, joined by Microsoft's venture fund M12 and Morgan Stanley. The headline use of the money is expanding the ability to watch AI agents that write and ship code while they are running, and to stop them mid-action.

※1 USD = 156 JPY (as of early September 2026)

Total Funding Passes 155 Million USD

HiddenLayer was founded in 2022 and is based in Austin, Texas. With this round, total funding passes 155 million USD (about 24.2 billion yen). The previous Series A was 50 million USD (about 7.8 billion yen), so a single round has now added more than everything raised before it.

The investor list has a distinct shape. Alongside lead investor Delta-v Capital are Booz Allen Ventures, Microsoft's M12, Morgan Stanley, and Ten Eleven Ventures. A consulting-linked fund with deep defense and intelligence ties, a major bank, and a cloud provider's investment arm sitting in the same line says a lot about where the company's customers are.

Co-founder and CEO Chris Sestito said the company set out to make secure enterprise AI possible well before most organizations felt the urgency, and that the round lets it keep growing the team and platform as agentic AI becomes core to how enterprises operate.

The Thing Being Protected Is the AI Itself

The platform covers a full set of functions for defending AI-based systems: discovery, which maps where models and agents actually exist inside an organization; AI supply chain security, which inspects the paths that external models and data travel; attack simulation, which probes for weaknesses by imitating real attacks; and runtime protection, which watches behavior in production.

The scope is not limited to generative AI. It spans the lifecycle of predictive models and agents as well, and includes protecting the models themselves and preventing intellectual property leakage. Where conventional security products look at applications and networks, HiddenLayer positions itself on the model and agent layer.

Coding Agents Sat Outside the Monitoring

The specific area named for expansion is protection of AI coding agents. On August 3, the company released Agent Harness Security, which extends the runtime protection module.

The feature plugs directly into the native hook surface each coding agent exposes, meaning the interception points built into the agent's own workflow. It detects and blocks prompt injection, sensitive data exposure, and unsafe command execution as they happen. It can redact sensitive information before a model sees it, steer an agent away from poisoned tool responses, and stop dangerous operations before they run.

From a security team's point of view, prompts, tool calls, shell commands, file edits, and repository interactions all become visible in one place. The more code that gets written and shipped without a human review step, the more a team needs to reconstruct what happened after the fact. Gartner predicts that by 2028, 90 percent of enterprise software engineers will use AI code assistants, up from less than 14 percent in early 2024. The surface to be monitored is set to widen sharply within a few years.

The Early Numbers Are Public

Some business figures have been disclosed. Annual recurring revenue grew more than 10 times over the past 12 months, and net new customers exceeded 50. The named segments are banking, insurance, pharmaceuticals, airlines, and the US defense and intelligence communities.

On the technical side, the company holds 39 granted patents and has 65 pending, spanning adversarial detection, model protection, and AI threat analysis. That is a thick stack for a company founded in 2022, and it points to sustained investment in research staff.

Valuation and absolute revenue figures were not disclosed. A 10 times increase is also easier to reach from a small base. A firmer read will have to wait for renewal rates once agentic AI deployment is genuinely widespread.

Runtime Is Becoming the Main Battleground

AI security has so far leaned toward pre-deployment inspection: check the model for problems, check the training data for poisoning, confirm everything before release. Agents break that pattern. They use the tools handed to them on their own judgment and decide their next action based on strings received from the environment. Inspection alone cannot guarantee behavior once the thing is moving.

HiddenLayer directing money toward runtime visibility and blocking follows that structural change. Similar products are multiplying, and services that vet agent behavior or block it before publication have raised money one after another in recent months. On the point that the object of defense is shifting from the model to the agent's behavior itself, the industry has largely converged.

Summary

HiddenLayer raised 100 million USD in a Series B, pushing total funding past 155 million USD. Booz Allen Ventures and Microsoft's M12 are among the investors, and the money is aimed primarily at runtime protection for AI coding agents. Agent Harness Security, released in August, hooks into the agent's own workflow to stop prompt injection and unsafe commands where they occur. Valuation and revenue remain undisclosed, but the shape of this round makes one thing clear: the main battleground in AI security is moving from pre-deployment inspection to monitoring during execution.