Share with your CISO
HiddenLayer is betting that runtime AI security, protecting AI agents as they operate rather than just at the design stage, is becoming a non-negotiable enterprise control. The company closed a $100m Series B led by Delta-v Capital, with Morgan Stanley, Microsoft’s M12, and Booz Allen Ventures participating. ARR grew more than tenfold over the past year, and the platform now covers customers across financial services, pharma, defense, and government. A newly launched Agent Harness Security product extends protection specifically to autonomous coding agents writing and deploying code with limited human review.
What this means for your business
The customer list here is the tell. Securities brokerages, defense and intelligence agencies, and a frontier model provider securing a platform used by 700 million people weekly don’t sign enterprise security contracts on speculation. If your organization is already running AI agents in production, whether for code generation, customer interactions, or back-office automation, you’re in the same exposure category as HiddenLayer’s existing customers, whether or not you’ve treated AI runtime as a security surface yet.
The distinction HiddenLayer is pressing, and the reason it matters to your vendor stack, is the gap between design-time and runtime security. Most organizations have applied guardrails to AI at the point of model selection or prompt design, the equivalent of checking a car’s brakes before the trip. Runtime security is the equivalent of monitoring the brakes while the car is moving. Agentic AI, meaning AI systems that take sequences of actions autonomously, like writing code, calling APIs, or executing workflows, creates an attack surface that static pre-deployment checks simply cannot cover. Adversarial prompt injection, where a malicious input hijacks an agent mid-task, is the canonical threat here, and it requires live detection to catch.
The tenfold ARR growth is impressive, but the more revealing number is 65 pending patents alongside 39 granted ones. That’s a company still in full research-and-productization mode, which means the platform will keep shifting under any enterprise contract. The CISO who signs with HiddenLayer today is also betting on a roadmap that isn’t finished. That’s not a disqualifier, it’s the realistic state of AI security broadly, but it means the renewal conversation in 18 months will look very different from the procurement conversation now. The right lens isn’t whether to buy runtime AI security; it’s whether you’re evaluating it as a standalone point solution or as a layer that needs to integrate with your broader SIEM and SOC infrastructure from day one.
Concept deep-dive: Adversarial prompt injection
Adversarial prompt injection is the AI-era equivalent of SQL injection: a malicious input, often hidden inside content an agent is asked to process, that overrides the agent’s intended instructions and redirects its behavior. Because agentic AI systems act autonomously across tools and systems, a single successful injection can cascade through multiple downstream actions before any human sees the result. This is why design-time guardrails alone are insufficient and why runtime monitoring, watching what the agent actually does, not just what it was told to do, is the emerging control category.
Based on reporting from HiddenLayer bags $100m to guard enterprise AI agents, originally published 2026-09-03 11:00:00.
