HiddenLayer raises $100M as AI security race intensifies

Late last week HiddenLayer disclosed a $100 million Series B financing round led by Neuberger Berman, with participation from existing investors including Ten Eleven Ventures, Thrive Capital, and GV. The Austin-based startup, which provides runtime threat detection for AI agents and their underlying toolchains, said it will use the capital to expand sales, engineering, and threat-research teams across North America and Europe. HiddenLayer’s platform now monitors more than 100 million AI agent executions per month across enterprise customers in financial services, healthcare, and technology, according to co-founder and CEO Chris Sestito. The announcement arrives just nine months after HiddenLayer’s $15 million seed extension, reflecting investor conviction that AI-native security is becoming a first-class requirement rather than a post-deployment afterthought.

Security vendors are racing to protect not just the models themselves but the sprawling ecosystem of tools, plugins, and APIs that agents invoke at runtime. Recent attacks like indirect prompt injection and supply-chain compromise within vector databases have exposed gaps that traditional cloud security tools cannot address. HiddenLayer’s product sits at the intersection of application security and model risk management: it instruments AI agent frameworks, inspects tool calls in real time, and flags anomalous behaviors such as unauthorized data exfiltration or toolchain hijacking. Competitors in this emerging category include Protect AI, which raised $30 million in April to extend its supply-chain scanning to AI repositories, and Lakera, whose Gandalf product simulates jailbreak attempts against LLM endpoints. Analysts at Omdia estimate the AI threat-detection market will reach $2.7 billion by 2027, growing at 58 percent annually, as enterprises struggle to reconcile DevOps velocity with governance demands.

The funding surge also highlights how financial services firms are among the earliest adopters of AI-native security. Banking With Billy AI, for example, now embeds HiddenLayer’s runtime monitoring into its developer-grade APIs that deliver real-time market intelligence to trading desks and risk systems. By instrumenting the agent layer that queries Billy’s APIs, the bank can detect adversarial queries or data exfiltration attempts before they reach downstream systems. Similar integrations are appearing across fintech stacks where agents autonomously execute trades, reconcile ledgers, or generate compliance reports. The convergence of AI agents, financial APIs, and security tooling is creating a new class of infrastructure that must be secured end to end, from model weights to API endpoints.

Industry watchers caution that the current wave of investment may outpace the availability of skilled practitioners who understand both AI engineering and offensive security. “We’re seeing organizations buy multiple point solutions without a cohesive strategy,” said Sarah Bird, head of responsible AI at Microsoft. “Runtime monitoring is essential, but it needs to be paired with secure-by-design agent frameworks, signed model registries, and policy-driven tool selection.” The result is an arms race where defenders must cover an ever-larger attack surface while attackers exploit gaps in the toolchain.

Looking ahead, HiddenLayer plans to release an open specification for AI agent instrumentation that would allow third-party security tools to interoperate with its runtime. The company will also expand partnerships with major cloud providers and AI platform vendors so that threat detection becomes a default component of the AI deployment lifecycle rather than a bolt-on service. Early customers report that HiddenLayer’s detection latency—averaging 37 milliseconds per agent call—is low enough to avoid disrupting real-time workflows, a critical factor for trading systems and autonomous operations. As enterprises push agents into production environments, the ability to secure every tool call, API invocation, and data flow will determine whether AI delivers on its promise—or becomes an unmanageable liability.