HiddenLayer Raises $100 Million for AI Runtime Security

What happened

HiddenLayer, the Austin-based AI security company, has raised $100 million in new funding to expand its AI runtime security platform, with a stated focus on securing agentic AI systems and the coding agents that have grown rapidly as development tools over the past year. The round marks a significant bet by investors that runtime security — monitoring and protecting AI models as they execute, not just before deployment — is becoming a distinct and commercially viable category.

This piece is for security engineers, platform builders, and engineering leaders who are evaluating runtime protection for AI systems, particularly those already running or planning to run AI coding agents in production. It covers what HiddenLayer is building with the capital, why the timing matters for the agentic AI moment, and what questions a technical buyer should still press the company to answer before adopting its products.

How it works

HiddenLayer, headquartered in Austin, Texas, raised $100 million in this round. The company builds AI security products focused on runtime protection — meaning it monitors AI models during inference rather than relying solely on pre-deployment scans or red-teaming. SecurityWeek covered the announcement, though the company has not publicly disclosed the identities of all investors participating in this round or the post-money valuation. Builders evaluating the company should verify whether the round is structured as equity, debt, or a mix, and whether any strategic investors hold board seats that would influence product direction. The size of the round — $100 million — puts HiddenLayer in a small tier of security startups that have raised at this scale specifically targeting AI infrastructure rather than traditional application or network security.

Why it matters

The timing corresponds closely with the acceleration of agentic AI adoption in software development. AI coding agents — tools that autonomously write, review, and execute code — are moving from experimental to production use inside engineering organizations, and they introduce attack surfaces that conventional security tooling was not designed to address. An agent that can read a repository, propose a pull request, and trigger a deployment pipeline can also be manipulated through prompt injection, adversarial inputs, or model tampering in ways that bypass code review and SAST tools entirely. Investors appear to be pricing in the expectation that enterprises deploying these agents will need a dedicated runtime layer and will pay for it as a distinct security control rather than an add-on to existing platforms.

HiddenLayer has stated it will invest the $100 million specifically in agentic runtime security capabilities. That framing suggests the capital is earmarked for product development aimed at protecting AI coding agents as they execute — detecting manipulation, monitoring model behavior in real time, and providing controls that can interrupt or audit an agent mid-task. The shift from securing static models to securing agents that take multi-step autonomous actions requires different detection primitives: agents do not just return a single output, they chain tool calls, access external systems, and produce side effects. Building detection logic for that behavioral surface is meaningfully harder than scanning model weights for known vulnerabilities. Builders should ask HiddenLayer specifically which integration points exist today — whether coverage is at the model layer, the orchestration layer, or the tool-call boundary.

Who is affected

Runtime security for AI is a small but quickly crowding field. Several vendors, including established security companies that have added AI model scanning to existing platforms, are competing for the same enterprise budget. HiddenLayer's positioning distinguishes it by focusing on the inference-time attack surface rather than the training or supply chain surface, which is where most AI security tooling has concentrated so far. Agentic security — specifically protecting coding agents — is a narrower and more recent positioning that few vendors have claimed with dedicated products. A builder comparing vendors should press each on whether agentic coverage is a deployed product, a roadmap item, or a marketing reframe of existing behavior-monitoring features. The $100 million gives HiddenLayer the resources to staff specialized research and build dedicated detection models, which smaller competitors with less capital will find difficult to match at the same pace.

What to watch next

Several material questions remain unanswered after this announcement. HiddenLayer has not published technical documentation on how its agentic runtime monitoring integrates with specific coding agent frameworks — the mechanisms by which it instruments agent execution, what telemetry it collects, and what latency overhead that monitoring introduces are all unspecified publicly. A production engineering team needs to know whether the product works with their orchestration layer without requiring significant architectural changes. Second, the company has not disclosed what threat models it explicitly covers for coding agents: prompt injection delivered through a code review comment, exfiltration through a model's tool calls, and adversarial manipulation of agent memory are meaningfully different problems, and buyers should confirm which of those HiddenLayer's current product addresses versus which are on the roadmap. The $100 million makes the build-versus-buy calculation more favorable to HiddenLayer's customers, but only if the product surface matches the actual threat surface those customers face.