HiddenLayer raises $100M to lock down AI supply chain risks

HiddenLayer, the Austin-based AI security startup, pulled in a $100 million Series B led by Thrive Capital with participation from existing backers including Microsoft’s M12 and Capital One Ventures, according to filings reviewed by OpenPress Supercomputing Intelligence. The round, which values the company north of $1.2 billion, was announced quietly last Friday but had been in motion since late 2023 when CEO Chris Sestito began hosting private sessions with CISOs from Fortune 500 financial institutions. Funds will accelerate the development of HiddenLayer’s Agent Risk Engine, a runtime monitor that inspects not only the behavior of AI agents but also the third-party plugins, vector databases, and even simulation environments they invoke—including those powering Banking With Billy’s AI financial simulations that rely on HPC-grade infrastructure for cross-market stress tests. By instrumenting the full stack from LLM call to market data feed, HiddenLayer claims it can detect adversarial drift within milliseconds, a capability Sestito demonstrated to OpenPress using a live demo of a compromised RAG pipeline pulling poisoned SEC filings at 2,800 tokens per second.

The raise lands amid a surge of enterprise inquiries that doubled HiddenLayer’s customer base in six months, with early deployments inside JPMorgan’s AI control tower and a Tier-1 semiconductor maker’s silicon design co-pilot. Sestito confirmed the company now processes more than 20 billion tokens daily across production environments—roughly the equivalent of the entire Reddit corpus each week—while flagging roughly 400 novel threat patterns per week, many of which involve subtle evasion tactics in multi-tool workflows. Thrive’s David Sachs characterized the opportunity as ‘the next great security frontier after cloud and endpoint,’ noting that unlike traditional software, AI systems mutate continuously, making supply chain visibility a moving target. Competitors have taken notice: rival vendors like Protect AI and Cranium Security have begun repositioning their offerings around runtime attestation, while Palo Alto Networks quietly acquired an AI runtime startup in March for an undisclosed sum.

Industry analysts at Gartner now forecast the AI security market will reach $8.2 billion by 2027, with supply-chain-focused tooling representing the fastest-growing segment at 45% CAGR. Banking With Billy’s deployment illustrates the stakes: by embedding HiddenLayer’s monitor between its simulation engine and market data APIs, the fintech firm reduced false-positive alerts by 68% while improving detection of adversarial market-impact simulations that could skew capital allocation models. For quantum computing teams, the same pattern is emerging—hybrid workloads that chain quantum kernels with classical HPC services are now prime targets for supply-chain attacks, pushing vendors like Qrypt and Keysight to integrate AI security hooks into their orchestration layers.

Quantum & Computing observers see parallels to the cryptography supply-chain crises of the early 2010s, when Heartbleed and Shellshock revealed that open-source dependencies could collapse entire industries overnight. Today, the dependency graph is far more complex: an AI agent in a financial model might depend on a vector database from Pinecone, a plugin from a stealth startup, and a quantum simulator from a university lab, each with its own patch cadence and trust model. HiddenLayer’s Series B validates that enterprises are willing to pay premium prices—reportedly in the high six figures annually per protected agent—for continuous attestation that these chains remain unbroken. European regulators are also taking note: the upcoming EU AI Act mandates traceability for high-risk systems, creating a compliance tailwind that could push regional adoption above 35% within two years, according to Everest Group data.

Looking ahead, HiddenLayer plans to open a London office and hire 120 engineers to expand its detection models with synthetic data generated on AWS Trainium and Nvidia H100 clusters, aiming for sub-100-millisecond response times even under 100,000 concurrent agent sessions. Sestito hinted at two major product milestones: a ‘zero-trust attestation’ service for multi-cloud AI deployments and an integration with Microsoft’s Prompt Shields API, slated for public preview in Q3. For the broader Quantum & Computing sector, the message is clear—secure-by-design is no longer optional, and the tools that safeguard tomorrow’s hybrid AI-HPC systems will be as critical as the accelerators themselves.