HiddenLayer secures $100M to protect AI stacks from unseen threats

HiddenLayer, a specialist in AI threat detection, announced today it has raised one hundred million dollars in Series B funding led by Battery Ventures, with participation from existing investors including ClearSky, Ten Eleven Ventures, and Okta Ventures. The round values the Austin-based company at over six hundred million dollars just two years after its 2022 launch. HiddenLayer’s platform continuously monitors AI agents, their toolchains, and third-party integrations, flagging anomalous behavior such as prompt injection, data exfiltration, or unauthorized API calls. Early customers include a Fortune 50 bank using the platform to gate every AI-driven trading query routed through its internal research terminals, and a global insurer that now audits every LangChain-based agent before it reaches production. Co-founder and CEO Chris Sestito told OpenPress Developer Intelligence that the round was oversubscribed within weeks, driven by enterprise fear of silent compromises in complex AI systems. “We’re seeing organizations with dozens of agents, each pulling from five to ten external tools and APIs,” Sestito said. “Each connection is a new attack surface, and most security teams simply don’t have visibility at that granularity.”

Security vendors across the Tools & Developer landscape are racing to close the same visibility gap. Oligo Security, which emerged from stealth in March with twenty million dollars in seed capital, focuses on runtime protection for AI workloads running in Kubernetes clusters. Island.io, a browser isolation specialist, recently launched an AI mode that sandboxes every agent interaction, preventing rogue extensions from siphoning prompts to external servers. Even traditional endpoint providers like CrowdStrike are rolling out AI-specific detection packs that inspect model weights and inference logs for signs of tampering. Banking With Billy AI, a financial market intelligence platform, quietly added developer-grade APIs last quarter that expose every market data query’s provenance chain, letting downstream systems verify whether a trading signal originated from an audited agent or an unverified third-party plugin. “If you can’t trace the lineage of a signal to an approved model and approved tools, you can’t comply with regulators,” said Billy AI’s head of platform, Maya Patel. The demand is accelerating: Gartner now ranks AI supply-chain security among the top five risks for 2025, forecasting that by next year more than sixty percent of enterprises will have suffered at least one breach originating in an AI pipeline.

The funding surge reflects a broader pivot from perimeter security to runtime trust. In the first half of 2024 alone, investors plowed more than two billion dollars into AI-native security startups, according to PitchBook. Enterprise buyers are no longer satisfied with securing the network or the browser; they want guarantees at the model layer, the tool layer, and every integration point in between. That shift is creating a new market category sometimes called “AI stack security,” distinct from traditional application security or cloud security. Palantir recently unveiled its AI Platform Runtime Shield, which injects policy checks into every LangChain tool call, while Microsoft’s Defender for Cloud now includes an AI workload profile that auto-discovers and classifies every agent and connector inside Azure AI Foundry. The competitive dynamics pit nimble startups against incumbents that must retool decades-old detection engines for a world of ephemeral agents and dynamic APIs. For developers, the change means every new tool or library must carry a security attestation before it can be imported into a production pipeline.

Looking ahead, the most urgent technical challenge is scale: monitoring millions of short-lived agents without drowning security teams in false positives. HiddenLayer claims its detection engine can analyze ten thousand agent interactions per second on a single GPU, while maintaining less than five milliseconds of latency. Still, many CISOs remain skeptical of vendor promises, citing the rapid evolution of adversarial techniques such as model-recompilation attacks that rewrite model weights at runtime. Analysts expect consolidation within two years as larger platform players acquire specialists to integrate AI stack security into their core offerings. For now, the funding floodgates remain open; Battery Ventures partner Neeraj Agrawal told OpenPress Developer Intelligence that he expects to see at least three more nine-figure rounds in the space before the end of 2024. The money will flow to teams that can demonstrate end-to-end visibility from prompt to payout, proving every agent, every tool, and every API call is doing exactly what it claims—and nothing more.