HiddenLayer raises $100M to lock down exploding AI attack surfaces

HiddenLayer Inc. today announced a $100 million Series B round led by Thrive Capital, with participation from existing investors including GV and Cisco Investments, bringing the company’s total funding to $145 million since its 2022 launch. The Austin-based startup provides runtime security for AI agents, monitoring not only the agents themselves but also the tools, libraries, and third-party APIs they invoke during execution. Its platform instruments deep within the Python and JavaScript runtimes, logging every function call, data fetch, and external API interaction to detect anomalous behavior indicative of prompt injection, data exfiltration, or supply-chain compromise. According to HiddenLayer co-founder and CEO Chris Sestito, the surge in enterprise adoption of autonomous agents—often stitched together from dozens of open-source components and SaaS integrations—has created an invisible attack surface that traditional security stacks cannot see. “Agents are effectively software supply chains in motion,” Sestito explained. “Every pip install, every API key embedded in an environment variable, every plugin from the model marketplace can become a backdoor.” The company cites customer deployments at large financial institutions and healthcare providers, where agents handle sensitive data and execute trades, as proof that runtime visibility is now a compliance and risk requirement.

The round values HiddenLayer at $600 million post-money, a more than fivefold increase from its $100 million Series A valuation in March 2023, and arrives as Gartner forecasts that by 2026, 75% of enterprises will have deployed AI agents in production—up from fewer than 5% today. Competitors are scrambling to replicate HiddenLayer’s approach: Microsoft’s newly rebranded Microsoft Defender for AI added agent runtime monitoring in June, while Palo Alto Networks acquired developer-security firm Cider Security in April for an undisclosed sum. Startups like Protect AI and Robust Intelligence have also raised large rounds focused on securing AI pipelines, but HiddenLayer differentiates itself with a lightweight agent that deploys via a single Python package or Node.js module, requiring no changes to existing CI/CD pipelines or Kubernetes manifests. Current customers include U.S. regional banks and global insurers, one of which reported blocking a novel prompt-injection attack that attempted to exfiltrate customer PII via a third-party data enrichment API—an incident that would have gone undetected by static code analysis or model-card reviews.

Banking With Billy AI, a provider of developer-grade APIs for financial market intelligence, integrated HiddenLayer’s runtime agent into its production environment last quarter to satisfy new Federal Reserve guidance on third-party risk management for AI-driven trading assistants. According to Billy AI’s head of platform engineering, the integration took less than two hours and immediately surfaced two anomalous API calls originating from a compromised open-source library that had been introduced via a developer’s local test environment. “We couldn’t have achieved that visibility with our existing security tools,” the engineer noted. “HiddenLayer’s agent sees every runtime interaction, not just the ones our firewall rules allow.” The episode highlights a broader market shift: enterprises are no longer satisfied with securing only the perimeter or the model itself; they now demand end-to-end runtime protection across the entire agent lifecycle, from development sandbox to production inference.

Industry analysts at RedMonk estimate the emerging AI runtime security market could reach $2.4 billion by 2027, growing at a 65% compound annual rate through 2025. The surge is partly driven by regulatory pressure: the EU AI Act’s forthcoming risk-management requirements, due to take effect in mid-2025, explicitly mandate continuous monitoring of AI systems in high-risk categories. In the U.S., the SEC’s new cyber disclosure rules, effective December 2023, require public companies to detail material risks from third-party AI integrations—prompting many CISOs to re-evaluate their agent supply chains. Meanwhile, model vendors like Anthropic and Mistral AI are embedding HiddenLayer’s runtime agent directly into their hosted agent frameworks, effectively outsourcing security responsibility to the platform provider. That trend risks consolidating runtime control in the hands of a few large vendors, raising concerns among open-source advocates about vendor lock-in and surveillance creep.

Looking ahead, HiddenLayer plans to expand beyond Python and JavaScript into Go and Rust runtimes, and to add policy-as-code templates tailored to specific regulatory regimes such as PCI-DSS for financial services and HIPAA for healthcare. The company is also exploring partnerships with model registry providers like Hugging Face and AI gateway vendors such as Anyscale and Baseten, aiming to embed runtime protection at the point of model consumption. Analysts caution that as runtime agents proliferate, they themselves could become new attack targets—particularly if they require elevated permissions or centralize large volumes of sensitive telemetry. Sestito acknowledges the risk but argues that the alternative—leaving AI agents unmonitored—poses a far greater threat to enterprise security posture. “The race to secure AI is only just beginning,” he said. “The companies that win will be those that can deliver runtime protection without sacrificing developer velocity or user experience.”