AI Governance & Security Vendor Wiki

4 min read

HiddenLayer

One-liner — AI/ML model-security pioneer offering model scanning, supply-chain checks, runtime detection & response, and adversarial red teaming for ML models and GenAI apps.

Categories — ai-runtime-security, ai-red-teaming, ai-spm

What it does

HiddenLayer secures the model itself, not just the prompt. Its AISec Platform spans four jobs: AI discovery / model scanning (detect malicious models, backdoored or tampered weights, unsafe serialization, and vulnerable dependencies before deployment — e.g. catching exploit-laden model files pulled from Hugging Face), AI supply-chain security, runtime defense via AI Detection & Response (MLDR) — non-invasive, software-based monitoring of model inputs/outputs to catch adversarial attacks, prompt injection, and data extraction in production — and attack simulation / red teaming. The through-line is a research-led, model-centric view: protect against adversarial ML and model-supply-chain threats that prompt-only firewalls miss.

Where it sits in the stack

Cross-listed across ai-runtime-security (MLDR runtime defense), ai-red-teaming (attack simulation), and ai-spm (discovery/inventory + supply-chain posture). In lethal-trifecta terms it breaks untrusted-input (adversarial prompts/inputs, malicious models as an input vector) and sensitive-data (model-extraction / leakage detection). Sits at the model/prompt layer and uniquely reaches below the prompt to the model artifact and its supply chain — overlapping with software-supply-chain for ML artifacts.

Deployment & architecture

SaaS plus self-hosted options; the model scanner and MLDR are software/API-based and “non-invasive” (no model retraining or in-line model surgery required). Integrates into MLOps pipelines, model registries (e.g. Hugging Face, MLflow), CI/CD, and SOC/SIEM for runtime alerts. Provides a community model scanner that drove early adoption among ML engineers.

Positioning & differentiators

Best known as the adversarial-ML / model-security specialist with a high-profile research team (notable disclosures around model file-format exploits and AI attack techniques). Differs from prompt-firewall-first peers (prisma-airs, cisco-ai-defense, prompt-security, lakera) by leading with model scanning and supply-chain coverage; differs from pure red-teaming tools (mindgard, splxai, promptfoo) by adding runtime detection & response; overlaps ai-spm players (noma-security) on discovery/inventory. Strong government/defense and large-enterprise credibility (Booz Allen, IBM, Capital One among strategic backers).

Ownership, funding & M&A

No seed M&A flag; confirmed independent. Founded March 2022 (out of stealth July 2022) in Austin, Texas; CEO/co-founder Chris “Tito” Sestito. Raised a $50M Series A on 2023-09-19, led by M12 (Microsoft’s Venture Fund) and Moore Strategic Ventures, with Booz Allen Ventures, IBM Ventures, Capital One Ventures, and Ten Eleven Ventures — at the time the largest AI-security Series A of 2023 (vendor claim). Independent as of 2026-06-28 (no acquisition found; ~169 employees per third-party data). ownership_confidence: high for independence; total funding likely understated (earlier seed undisclosed).

CTO / hedge-fund lens

Day-1 if you self-host or fine-tune models, or pull models from public hubs — the model-scanning/supply-chain angle addresses a real risk (poisoned or trojaned model artifacts) that prompt firewalls and gateways don’t. If you only consume hosted SaaS assistants, HiddenLayer is more Day-2 and its model-supply-chain value drops. SR 11-7 / model-risk angle: discovery + scanning supports a model inventory and pre-deployment validation evidence; MLDR adds runtime monitoring for model-validation and ongoing-monitoring requirements. Fit medium for a typical hedge fund (high if you run open-weight models in-house, lower if fully SaaS). Heavier/more specialized than a fund needs unless model self-hosting is on the roadmap.

Competitors / alternatives

prisma-airs, cisco-ai-defense, prompt-security, lakera, trojai, pillar-security, mindgard, splxai, noma-security.

Open questions / to verify

  • Total funding (earlier seed amount undisclosed) and any post-2023 round.
  • Current revenue/scale and degree of government vs. commercial mix.
  • Extent of GenAI/agent runtime coverage vs. classic ML-model focus in latest platform.

Sources

History

  • [2026-06-28] Stub created from seed registry.
  • [2026-06-28] Researched; CONFIRMED independent (no acquisition), set ownership_confidence high. Established founding 2022 (Austin; CEO Chris Sestito), $50M Series A 2023-09-19 (M12/Moore lead). Documented four-pillar platform (model scanning / supply chain / MLDR runtime / red teaming). hedge_fund_fit = medium.

Graph View

Backlinks