HiddenLayer announced a $100 million Series B on September 2, 2026, led by Delta-v Capital. Microsoft's venture arm M12, Morgan Stanley, Ten Eleven Ventures, and Booz Allen Ventures also participated. The company scans machine learning model files for backdoors, serialization exploits, and adversarial perturbations across more than 35 file formats.
Why this matters more than a standard funding round
The AI security vendor landscape has consolidated significantly over the past twelve months. Lakera, which built the best prompt injection detection technology, is now part of Check Point. Protect AI, Robust Intelligence, CalypsoAI, and Prompt Security were all acquired between October 2024 and September 2025. HiddenLayer raising $100 million as an independent company signals that at least some enterprise buyers see model security as a standalone platform problem, not a feature to bundle into an existing security product.
What HiddenLayer actually does
The platform addresses three distinct threat categories. First, model supply chain threats: backdoored models distributed via public repositories, malicious serialization payloads embedded in pickle or ONNX files, and model extraction attacks where an adversary reconstructs a proprietary model through repeated inference. Second, adversarial robustness: testing whether deployed models can be manipulated by adversarial inputs to produce attacker-controlled outputs. Third, AI asset discovery: finding all models running in an enterprise environment, which is harder than it sounds because models get embedded in applications without security teams knowing.
The threat model practitioners need
MITRE ATLAS catalogs 64 attack techniques against AI systems. The ones operationalized today are model poisoning during fine-tuning (relevant to any organization that fine-tunes foundation models on proprietary data), serialization exploits (the pickle format used by PyTorch is effectively arbitrary code execution if the model file comes from an untrusted source), and prompt injection in agentic systems (the number one risk in the OWASP LLM Top 10, with an 84% success rate in current agentic system tests). HiddenLayer's tooling addresses the first two directly; prompt injection is a separate product category.
What the consolidation means for buyers
With most independent AI security startups now acquired, enterprise buyers have two practical options: use the AI security module bundled inside their existing platform, or use HiddenLayer as the remaining independent specialist. The consolidation increases competitive pressure on HiddenLayer to expand its platform or find an acquirer. Booz Allen Ventures participating is worth noting: Booz Allen is a major US defense and intelligence contractor, suggesting HiddenLayer has federal customer traction that the commercial narrative underplays.
Gigia Tsiklauri is a Security Architect and founder of Infosec.ge. Get in touch if your organization is building or deploying AI systems and needs help mapping your model security posture.