What Is HiddenLayer? $100M Funding and the Rise of AI Agent Runtime Security
Enterprises are quickly deploying autonomous AI agents that write code, invoke tools, access data, and execute multi-step workflows with minimal human oversight. These systems expand the attack surface far beyond traditional models, creating urgent demand for specialized runtime defenses that monitor behavior as it occurs. HiddenLayer, an Austin-based AI security company founded in 2022, has positioned itself at the center of this shift by securing agentic, generative, and predictive AI applications across the full lifecycle. On September 2, 2026, the company announced a $100 million Series B funding round led by Delta-v Capital, with participation from Ten Eleven Ventures, Morgan Stanley, Microsoft’s M12, and Booz Allen Ventures, bringing total capital raised to approximately $156 million.
The funding arrives as annual recurring revenue grew more than 10 times over the prior year into the tens of millions of dollars, driven largely by more than 50 new platform customers across financial services, technology, pharmaceuticals, airlines, government, and U.S. defense and intelligence communities. One customer is a leading frontier model provider supporting more than 700 million weekly users. This article examines HiddenLayer’s platform evolution, the technical and market drivers behind agentic runtime security, verified growth metrics, and the practical implications for organizations deploying AI agents in production. The central thesis is that purpose-built runtime visibility, detection, and enforcement for AI agents have become an essential infrastructure as autonomous systems move from experimentation into core enterprise operations.
HiddenLayer announced its $100 million Series B on September 2, 2026, with Delta-v Capital leading and strategic participation from Ten Eleven Ventures, Morgan Stanley, Microsoft’s M12 venture fund, and Booz Allen Ventures. The round follows a $50 million Series A in September 2023 co-led by M12 and Moore Strategic Ventures and an earlier seed round, resulting in total funding of roughly $156 million. Company leadership stated the capital will deepen the enterprise platform, particularly Agentic Runtime Security and the recently introduced Agent Harness Security solution designed for AI coding agents. Annual recurring revenue expanded more than tenfold in the preceding twelve months, with over 90 percent of that growth attributed to new customers. Verticals with the strongest traction include financial services and large technology firms building AI products, alongside contracts supporting the Department of Defense and the intelligence community.
The raise reflects broader market recognition that AI systems require dedicated security layers rather than extensions of conventional tools. HiddenLayer’s researchers hold 39 granted patents and 65 pending patents covering adversarial detection, model protection, and threat analysis. The team has contributed to efforts involving CISA/JCDC, MITRE, NIST, OWASP, and OpenSSF. Leadership also strengthened commercial capacity by appointing a new chief revenue officer and signaled plans for further hiring, channel expansion, and entry into European and EMEA markets. These steps position the company to meet rising enterprise demand as agentic systems become operational infrastructure rather than pilot projects.
Agentic runtime security focuses on monitoring how AI agents behave while executing tasks in production environments. Traditional model defenses often concentrate on training data integrity, input filtering, or pre-deployment testing. Runtime approaches instead observe live interactions across prompts, tool calls, data access, and multi-step decision sequences. HiddenLayer describes this capability as analogous to endpoint detection and response systems, adapted specifically for AI inference and agent workflows. The platform reconstructs agent sessions to show how decisions unfold, enabling security teams to detect prompt injection, malicious tool sequences, data exfiltration, and unauthorized actions before they cascade.
According to HiddenLayer’s 2026 AI Threat Landscape Report, one in eight AI breaches is linked to agentic systems, yet nearly one-third of organizations cannot determine whether they have experienced such an incident. Agents operate at machine speed and can access systems, move data, or invoke tools far faster than human response cycles allow. Runtime security therefore emphasizes real-time visibility, investigation, threat hunting, and adaptive enforcement. Policies can redact sensitive information, block unsafe actions, or steer agents away from compromised inputs while preserving operational continuity. This continuous verification model addresses the reality that design-time safeguards alone cannot guarantee trustworthiness once agents interact with untrusted content or external tools.
On August 3, 2026, HiddenLayer introduced Agent Harness Security as an extension of its Runtime Security module. The solution targets AI coding agents that read source files, execute shell commands, install dependencies, modify repositories, and generate pull requests. These agents have evolved beyond simple suggestion tools; Gartner has projected that by 2028, 90 percent of enterprise software engineers will use AI code assistants, up from less than 14 percent in early 2024. The harness layer, the orchestration environment connecting models to context, memory, tools, and skills, creates a new runtime attack surface where indirect prompt injection embedded in a README or tool output can influence behavior.
Agent Harness Security integrates with each coding agent’s native hook surface to deliver inline visibility and control. Security teams gain insight into prompts, tool calls, file edits, repository interactions, and command execution. The system can detect secrets flowing into model context, unsafe command patterns, unexpected dependency installs, and obfuscated payloads. Enforcement options include redaction of sensitive data before it reaches the model, steering agents away from poisoned responses, and blocking high-risk actions. Because different agent platforms expose varying levels of control, the solution reports the actual enforcement strength available for each environment, allowing teams to understand what protections are active without disrupting developer productivity.
Customer adoption accelerated markedly in the year preceding the Series B. More than 50 new platform customers joined, spanning securities brokerage, banking, insurance, accounting, government, technology, IT services, pharmaceuticals, airlines, and U.S. defense and intelligence communities. International additions included a major pharmaceutical company, premium automotive brands, and food and beverage providers. One notable deployment supports a leading frontier model provider whose systems serve more than 700 million weekly users. Financial services and large technology organizations building AI products currently represent the largest commercial verticals.
Government validation further underscores the platform’s applicability in high-assurance settings. HiddenLayer was selected as an awardee on the Missile Defense Agency’s SHIELD IDIQ contract supporting air-gapped and classified environments. The company also participates in Department of Energy initiatives focused on secure AI applications. These contracts demonstrate that the same core technology, non-invasive analysis that does not require access to raw training data or model weights, can operate in both commercial production environments and sensitive national security contexts. The breadth of adoption indicates that organizations treating AI as core operational infrastructure are allocating dedicated security budgets rather than relying solely on existing tooling.
HiddenLayer’s platform includes AI supply chain security capabilities that examine models and their dependencies before deployment. The company reports scanning approximately 50 different AI file frameworks to verify that open-source and open-weight models match their claimed identity and integrity. Research has identified cases in which models purport to be one thing while containing hidden or substituted components. This analysis produces an AI Bill of Materials that evaluates model genealogy and flags malware, vulnerabilities, or integrity issues without requiring access to proprietary training data.
As enterprises increasingly incorporate third-party and open-weight models into production pipelines, supply chain verification becomes a foundational control. Attackers can introduce compromised models that appear legitimate yet contain backdoors or altered behavior. Runtime protections complement these pre-deployment checks by continuing to monitor behavior after models enter service. The combination addresses both static risks in the model artifact and dynamic risks that emerge only when the model interacts with live inputs, tools, and data sources. Organizations adopting agentic workflows that pull from multiple model sources benefit from continuous integrity validation across the lifecycle.
HiddenLayer maintains a dedicated adversarial AI research organization that has produced the first comprehensive Adversarial Prompt Engineering Taxonomy and identified dozens of vulnerabilities across foundation models, tools, and supporting infrastructure. The research portfolio includes 39 granted patents and 65 pending patents in areas such as adversarial detection, model protection, and threat analysis. Team members contribute to collaborative efforts with CISA/JCDC, MITRE, NIST, OWASP, and OpenSSF, helping refine red-teaming methodologies and inform discussions on AI governance and safety.
This research orientation differentiates the platform by grounding product capabilities in ongoing threat discovery rather than static rule sets. Findings feed directly into detection logic for prompt injection, agent manipulation, and malicious tool use. The same research underpins attack simulation modules that allow organizations to test defenses continuously. As agentic systems introduce new multi-step attack chains, the ability to update detection based on emerging research becomes a practical advantage for security teams responsible for production AI environments.
Gartner estimates that organizations will spend $2.83 billion in 2026 on products designed to secure AI tools, representing an 83 percent increase over 2025. The firm forecasts the market will reach nearly $4.78 billion in 2027. AI application security and AI usage control rank among the larger segments, with strong growth also projected for governance platforms and gateways. These figures reflect recognition that conventional security controls leave gaps when systems generate outputs, invoke tools, or act autonomously.
The broader context includes rising overall information security spending and the specific emergence of agent-related risks. Analysts note that more than half of successful cyberattacks on AI agents are expected to exploit access control weaknesses and prompt injections by 2029. Runtime protection has emerged as a leading category within AI agent security because it addresses live execution rather than solely pre-deployment testing or static policy. Enterprises scaling agentic workloads therefore confront both the productivity gains of autonomous systems and the corresponding need for specialized visibility and enforcement layers.
Core runtime capabilities focus on three interrelated risks: prompt injection and indirect attacks that manipulate agent behavior through untrusted content; sensitive data exposure when agents handle source code, credentials, or proprietary information; and unsafe agent actions that result from compromised reasoning or malicious tool responses. Detection operates across the full session context rather than isolated prompts, enabling identification of multi-step attack chains. Enforcement can block, redact, or shape content according to policy while integrating with existing SIEM and SOAR workflows.
For coding agents specifically, the harness-level integration observes file reads, command execution, dependency installation, and repository interactions. This level of observability allows security teams to investigate not only what an agent did but also how it reached that decision. In environments where agents operate with elevated privileges on developer machines or continuous integration systems, the ability to intervene before unsafe actions complete reduces the potential blast radius. Organizations report that preserving developer productivity remains essential; content-shaping techniques that allow agents to continue safely after redaction support that requirement.
Proceeds from the Series B are allocated primarily to deepening Agentic Runtime Security and Agent Harness capabilities, expanding engineering and research capacity, and accelerating sales and distribution. Leadership has indicated a focus on vertical scaling alongside artificial intelligence adoption while preparing for eventual horizontal expansion into related cybersecurity domains. International growth begins with Europe and the wider EMEA region, supported by additional hiring and channel development.
The capital also reinforces the company’s ability to maintain a purpose-built platform rather than relying on general-purpose security tools adapted after the fact. Investors highlighted that traditional security solutions were designed for code and infrastructure, not for models that can be poisoned, hijacked, or manipulated through their own inputs. The combination of research depth, non-invasive architecture, and runtime focus positions the platform for continued adoption as agentic systems move deeper into regulated and high-value environments.
Organizations introducing AI coding agents or autonomous workflow systems confront a visibility gap: many lack real-time inventories of active agents and cannot readily reconstruct multi-step sessions when anomalies occur. Implementing runtime security requires integration points that observe agent activity without introducing unacceptable latency or disrupting existing development pipelines. Solutions that leverage native agent hooks and report enforcement strength per platform help security teams understand residual risk.
Effective programs combine pre-deployment supply chain scanning, continuous attack simulation, and live runtime monitoring. Policies should address both direct prompt injection and indirect attacks delivered through tool outputs or retrieved content. Integration with existing security operations tooling enables detections to flow into established investigation and response processes. As agents gain privileges to modify code, access secrets, or interact with production systems, the cost of delayed detection rises, making continuous runtime controls a practical necessity rather than an optional enhancement.
The AI security market has attracted significant capital, with several vendors raising substantial rounds to address overlapping or adjacent problem spaces. Large cybersecurity firms have pursued acquisitions to add AI-specific capabilities. HiddenLayer differentiates through its focus on non-invasive analysis that does not require model weights or training data, its emphasis on agentic runtime and coding-agent harnesses, and its established presence in both commercial and classified environments. The platform unifies discovery, supply chain security, attack simulation, and runtime protection within a single architecture.
As the category matures, enterprises evaluate solutions based on coverage of multi-step agent workflows, ease of integration with existing agent frameworks, quality of session reconstruction for investigation, and the ability to enforce policy without excessive false positives or developer friction. Research-backed detection updates and contributions to industry standards provide additional signals of long-term viability. Organizations selecting platforms increasingly prioritize those that treat runtime behavior as a first-class control surface rather than an afterthought to model-centric defenses.
Agentic systems are transitioning from experimental tools to components of core business processes. Gartner and other analysts project continued quick growth in both AI agent deployment and corresponding security spend. Runtime security is expected to remain a primary focus because agents introduce dynamic, context-dependent risks that static controls cannot fully address. Capabilities that provide visibility across tools, data, and multi-agent interactions will likely expand as organizations increase the autonomy granted to these systems.
HiddenLayer’s recent funding and product extensions position the company to support that trajectory. The combination of documented ARR growth, customer diversity across regulated industries, government contracts, and a substantial patent portfolio indicates sustained demand for purpose-built AI security. Enterprises that treat runtime protection as essential infrastructure rather than an optional layer will be better equipped to scale agentic workflows while managing the expanded attack surface those workflows create. Continuous verification at runtime remains the practical foundation for trustworthy AI operations.
What specific capabilities does HiddenLayer’s Agentic Runtime Security provide for production agents?
Agentic Runtime Security delivers real-time visibility into how AI agents interact with tools, data, and other systems during live execution. It reconstructs full session context so security teams can understand decision sequences, detect prompt injection and malicious tool use, and apply enforcement actions such as blocking, redaction, or content steering. The module integrates with existing security operations workflows and supports investigation and threat hunting across multi-step autonomous processes.
How does Agent Harness Security protect AI coding agents without slowing developers?
Agent Harness Security integrates directly with the native control surfaces of supported coding agents. It observes prompts, file access, shell commands, dependency installs, and repository changes in real time. Sensitive data can be redacted before reaching the model, and agents can be steered away from poisoned inputs. Enforcement is calibrated to the capabilities of each agent platform, allowing productive work to continue while high-risk actions are constrained.
Which industries have adopted HiddenLayer’s platform most extensively?
Financial services and large technology companies building AI products currently form the largest commercial verticals. Additional strong adoption appears in insurance, pharmaceuticals, airlines, government, and U.S. defense and intelligence communities. The company also supports a frontier model provider serving more than 700 million weekly users and has international customers in the automotive and food and beverage sectors.
What does the $100 million Series B funding enable HiddenLayer to pursue next?
The capital supports deeper development of Agentic Runtime Security and Agent Harness capabilities, expansion of engineering and research teams, acceleration of sales and distribution, and geographic growth beginning with Europe and EMEA. Leadership has also indicated continued investment in the research organization that underpins detection updates and contributions to industry standards.
How does HiddenLayer analyze models without accessing training data or weights?
The platform uses non-invasive techniques that examine model artifacts, tensor structures, file frameworks, and runtime behavior. This approach allows scanning of approximately 50 AI file formats for integrity issues, malware, and mismatches between claimed and actual model content. The same philosophy extends to runtime monitoring, which observes inputs, outputs, and agent actions without requiring internal model access.
What market growth projections support investment in AI runtime security?
Gartner forecasts that spending on products to secure AI will reach $2.83 billion in 2026, an 83 percent increase from the prior year, and nearly $4.78 billion in 2027. Analysts expect access control weaknesses and prompt injections to feature in a majority of successful attacks on AI agents by 2029. These projections reflect the expanding attack surface created by autonomous systems in production environments.
If you worry about the frequent ups and downs in the market, and pursue a more stable option to earn money passively, KuCoin is the right place to come:
Simple Earn: Deposit and withdraw tokens anytime, earning stable returns.
Kucoin Earn: Earn stable profits with professional asset management.
Hold to Earn: Earn rewards by holding assets in Funding, Trading, Margin, Futures, Mining, and Unified Accounts.
Staking: Unlock the earning potential of on-chain assets.
Advanced Investments: Advanced Investments offer a variety of structured products to help your money grow in any market.
Shark Fin: Principal Protection and Guaranteed Gains
Dual Investment: Buy low and sell high with transparent return calculations.
Snowball: High yields, with price protection.
Discount Buy: Buy crypto at discount prices.
KCS Loyalty: Level up to enjoy exclusive perks by staking ≥ 1 KCS.
KuCoin Wealth: Discover future value and begin your smart investing journey.
KCS Benefits: Hold and stake KCS to access benefits across the platform.
KCS Staking 2.0: Participate in KCS on-chain governance to earn yield.
Disclaimer: This content is for informational purposes only and does not constitute investment advice. Cryptocurrency investments carry risk. Please do your own research (DYOR).