HiddenLayer Just Raised $100 Million. That Tells You Everything About Where AI Security Is Heading.
I’ll be honest: when I saw that HiddenLayer raised $100 million in Series B funding, my first thought wasn’t “great startup story.” It was: “This is the market telling us something we should have paid attention to a lot sooner.”
>HiddenLayer is an Austin-based AI security company. Their pitch is straightforward: they build tools to protect AI models, agents, and workflows from adversarial attacks, vulnerabilities, and malicious code injections. On September 2, 2026, they announced a $100 million Series B led by Delta-v Capital, with participation from Ten Eleven Ventures, Morgan Stanley, Microsoft’s M12 fund, and Booz Allen Ventures.
That’s a lot of money. But here’s the part that actually matters: the company’s annual recurring revenue grew more than 10x over the past year, they signed more than 50 new platform customers — including names in securities brokerage, banking, insurance, government, pharmaceuticals, airlines, and the US defense and intelligence communities — and one of their customers is reportedly a leading frontier model provider with more than 700 million weekly users.
When a company in a category that barely existed three years ago is growing at that pace, with that kind of customer list, it’s not a venture capital story. It’s a warning sign dressed up as a funding announcement.
The Market Data Behind the Headline
HiddenLayer doesn’t exist in a vacuum. Gartner projected in August 2026 that companies will spend $2.83 billion this year on products meant to secure AI tools — 83% more than in 2025 — and expects that number to reach nearly $4.78 billion in 2027. That’s a 68.7% increase year over year. For context, most enterprise software categories grow at single digits or low double digits. AI security is growing at venture-capital speeds while already being a multi-billion-dollar market.
The asymmetry that keeps coming up in the data is what should concern anyone deploying AI in production. Enterprises are spending roughly 17 times more on AI-powered security tools than on securing the AI itself, according to one analysis of Gartner’s forecast data. Think about that for a second: we’re building AI systems into production faster than we’re building defenses around them, and the spending gap between the two is an order of magnitude.
Gartner also predicts that 40% of enterprise applications will feature task-specific AI agents by the end of 2026, up from less than 5% in early 2025. Those aren’t chatbots. We’re talking about autonomous systems that scan network traffic, analyze logs, initiate responses, execute workflows — and do it at machine speed. When something goes wrong in one of those systems, there’s no “let me think about this for a minute” moment. The damage happens before a human can even open a ticket.
Why Runtime Security Is the Real Story
Here’s where I think the HiddenLayer story gets interesting, and where a lot of the coverage misses the point. The startup’s CEO, Chris Sestito, told TechCrunch that their biggest change over the past year wasn’t a pivot — it was an expansion of scope. They started with traditional machine learning model protection, added generative AI, and then had to extend everything to cover agentic AI systems.
That last part — agentic AI — is where the risk profile changes completely. AI agents need their own supply chain security — AIR raises $50M to build itAn AI model that generates text is one thing. An AI agent that can run shell commands, modify source code, access databases, call external APIs, and make autonomous decisions is something else entirely. HiddenLayer’s own research found that one in eight AI breaches are linked to agentic systems. Each agent interaction expands the operational blast radius.
Their new Agent Harness Security product, announced in August 2026, targets exactly this. It integrates into AI coding agents at runtime — the tools that developers are increasingly using to write, review, and ship code with less human oversight. It detects prompt injection embedded in source files, catches secrets flowing into AI tool calls, flags unsafe command execution, and blocks malicious dependency installs. The pitch is that it protects the development workflow without slowing developers down, which is the entire reason organizations adopt these tools in the first place.
This is the category that keeps me up at night as someone who manages an ICT division. We’re deploying AI assistants into production environments. We’re experimenting with agents that can execute tasks autonomously. And the security tooling to monitor what those agents are actually doing at runtime is still catching up to what the agents can do.
The Supply Side Problem Nobody’s Talking About
One of the things Sestito highlighted in the TechCrunch interview that I haven’t seen enough attention on: “We’re looking at things like models purporting to be one thing, but they’re another — hidden models inside of models.”
That’s a supply chain problem. If you’re pulling models from Hugging Face or any model repository, how do you know what you’re actually running? A model that claims to be a text classifier could contain hidden behavior that activates under specific conditions. A model fine-tuned for your specific use case could have been poisoned during the training process. The AI supply chain has all the same vulnerabilities as any other software supply chain — and we’ve spent the last several years failing to solve those at the traditional software level. how to audit your AI toolchain for supply chain risks
HiddenLayer’s platform covers discovery, runtime protection, attack simulation, and supply chain security. That last piece — AI supply chain security — is arguably the most underrated part of what they do. Most organizations have no inventory of the AI models they’re running, no process for vetting them before deployment, and no runtime visibility into what those models are doing in production. That’s a security posture that would get laughed out of any traditional IT audit.
The Elephant in the Room: Consolidation Is Coming
Here’s the uncomfortable question that follows any AI security startup raising $100 million: how long does this category exist as a standalone market?
Sestito acknowledged this directly in his interview. He said some parts of what HiddenLayer offers could eventually be bundled into platforms built by companies like Microsoft, OpenAI, and AWS. His bet is that AI infrastructure will grow toward governance features — discovery, identity, policy controls — rather than the runtime protection tools his company builds.
I think that’s partly right and partly optimistic. The big platforms absolutely will absorb some of this. Microsoft already has security products that touch AI. OpenAI and AWS will add guardrails to their own ecosystems. But the agentic AI runtime problem is genuinely new, and the big platforms have a structural incentive to make their own agents look safe rather than to provide neutral, cross-platform runtime monitoring. If your AI security strategy depends entirely on the vendor whose agents you’re trying to secure, you have a conflict of interest problem.
Meanwhile, the competitive landscape is heating up. Palo Alto Networks announced an intent to acquire Protect AI. Cisco is moving on AI security. Check Point acquired Lakera. Noma raised $100 million and Zenity raised $125 million for adjacent areas. This isn’t a category that’s about to disappear — it’s a category that’s about to get crowded, consolidated, and inevitably absorbed into the broader cybersecurity conversation.
What This Means for People Actually Running AI in Production
I want to be practical here, because there’s a temptation to treat every AI security funding announcement as a sign that “the industry is taking this seriously” and move on. That would be a mistake.
If you’re deploying AI in production — and the data says a lot of organizations are, whether they’ve formally acknowledged it or not — here’s what the HiddenLayer story should prompt you to think about:
Do you have an inventory of the AI systems you’re running? HiddenLayer’s platform includes AI discovery, which builds a living inventory including shadow AI. If you can’t answer this question, you have a gap that’s bigger than any single vendor can fill.
Can you see what your AI agents are doing at runtime? Not what they’re supposed to do. What they’re actually doing. Every tool call, every data access, every command execution. If your answer is “we trust the model” or “we have policies,” that’s not the same as runtime visibility. how to verify if a suspicious message is really from AmazonIt’s the difference between a firewall policy and a network monitor. You need both.
What’s your AI supply chain posture? Where do your models come from? How are they vetted? What happens if a model you’re running has been tampered with? This is the software supply chain problem we spent a decade not solving, now applied to a category that changes every month.
Who’s responsible when an AI agent does something it shouldn’t? HiddenLayer’s runtime enforcement can block an unsafe action in real time. But someone needs to define what “unsafe” means for your environment, and someone needs to be accountable when the enforcement layer misses something. That’s not a technology problem. That’s an organizational one.
Why the $100 Million Number Actually Matters
I don’t bring up the funding size to gawk at it. The number matters because it signals where the market thinks this category is headed. HiddenLayer’s 10x ARR growth and 50+ new customers — including defense and intelligence community contracts — tell you that organizations with real security requirements are buying into this category now, not waiting for it to mature. the DOJ just drew a line in the AI copyright war — and it’s not where you think
Morgan Stanley’s strategic investment arm participated in this round. That’s not a typical venture capital signal. When a major financial institution’s strategic investment team puts money into an AI security startup, they’re signaling that their own exposure to this risk is significant enough to warrant a stake in the solution.
Microsoft’s M12 fund was there too. That’s the venture arm of a company that sells both AI tools and security products — and even they see enough distance between those two businesses that investing in a third-party AI security player makes sense.
The Gartner numbers tell the same story from the other direction: $2.83 billion this year, nearly $4.8 billion next year. That’s not a niche. That’s a market that’s been waiting for the right moment to exist, and the moment has arrived.
Where I Land on This
I’m not here to tell you HiddenLayer is the answer to AI security. No single vendor is. The category is too new, the threat landscape is too fast-moving, and the attack surface is expanding faster than any one company can cover it.
But I am here to tell you that the category is real, it’s growing at a pace that dwarfs almost every other enterprise software segment, and the organizations buying into it — banks, defense contractors, pharmaceutical companies, frontier model providers — are not doing it because it’s a nice-to-have. They’re doing it because they’ve decided that deploying AI without runtime security is an unacceptable risk.
That should tell you something. If organizations with the most sophisticated security operations in the world are suddenly spending heavily on AI runtime protection Norway called them ‘pervert glasses.’ Here’s why, the rest of us should probably pay attention. The gap between what AI can do in production and what we can monitor and control at runtime is real, it’s widening, and pretending it doesn’t exist because the headlines are about funding rounds rather than breaches doesn’t make it go away.
HiddenLayer raised $100 million. The real story isn’t the money. It’s that the market finally decided AI security is a category worth paying for — and that decision is probably a few years later than it should have been.
FAQ: HiddenLayer and the AI Security Market
What is HiddenLayer?
HiddenLayer is an Austin-based AI security company that builds a platform to protect AI models, agents, and workflows from adversarial attacks, vulnerabilities, and malicious code injections. Their platform covers AI discovery, runtime protection, attack simulation, supply chain security, and agentic/MCP protection. As of September 2026, they serve over 50 enterprise customers including financial services, government, defense, and a leading frontier model provider.
Why did HiddenLayer raise $100 million?
The company’s annual recurring revenue grew more than 10x over the past year, driven by demand from enterprises deploying AI who need runtime security. The $100 million Series B, led by Delta-v Capital with participation from Morgan Stanley, Microsoft’s M12, Ten Eleven Ventures, and Booz Allen Ventures, will fund expansion of their agentic runtime security capabilities, international growth into Europe and EMEA, and continued R&D. The broader AI security market is projected by Gartner to reach $2.83 billion in 2026 and nearly $4.8 billion in 2027.
What is AI runtime security?
AI runtime security is the practice of monitoring and protecting AI systems — especially autonomous agents — while they’re running in production. Unlike traditional security checks that happen before deployment, runtime security watches what AI systems actually do in real time: every prompt, tool call, data access, and command execution. It can detect and block threats like prompt injection, malicious tool use, data exfiltration, and unsafe autonomous actions as they happen, rather than after damage is done.
How big is the AI security market?
Gartner projected in August 2026 that enterprises will spend $2.83 billion on AI security products in 2026, an 83% increase over 2025, growing to nearly $4.78 billion in 2027. For context, enterprises are spending roughly 17 times more on AI-powered security tools than on securing the AI systems themselves, according to analysis of Gartner’s forecast data. Gartner also predicts 40% of enterprise applications will include task-specific AI agents by the end of 2026.
What are the main AI security risks right now?
The biggest runtime risks include prompt injection (attackers manipulating AI systems through prompts, retrieved content, or tool outputs), indirect attacks where untrusted data influences agent behavior, model supply chain risks (models that appear to be one thing but contain hidden behavior), and agentic system risks where autonomous agents can access systems, move data, and take actions at machine speed. HiddenLayer’s 2026 AI Threat Landscape Report found that one in eight AI breaches are linked to agentic systems.
Is AI security a standalone market or will it be absorbed by major platforms?
Both. Major platforms like Microsoft, OpenAI, and AWS are adding AI guardrails and security features to their own ecosystems. But the cross-platform nature of AI deployments — organizations run models and agents from multiple vendors — creates a need for neutral, third-party runtime monitoring that platform-specific tools can’t fully cover. Competitors like Palo Alto Networks, Cisco, Check Point, Noma, and Zenity are all investing heavily in this space, suggesting the market will consolidate but not disappear into a single vendor’s offering.