Skip to content

Dreadnode

An offensive-security startup that red-teams AI models and agentic systems before and after deployment, founded by former NVIDIA and NetSPI researchers.

Visit Website ↗ + Add to Compare Claim This Company
57/100Incremental Innovator

Overview

Dreadnode builds offensive-security tooling purpose-built for AI systems rather than adapting traditional penetration-testing tools to LLMs. Its product line spans Crucible, a public AI hacking sandbox used to train and benchmark offensive AI agents; Strikes, an “AI agent training ground” where organizations deploy and evaluate offensive agents against realistic scenarios; and Spyglass, a red-teaming tool (in limited early access) for stress-testing AI applications already in production.

The company was co-founded by Will Pearce, formerly the AI red-team lead at NVIDIA, and Nick Landers, formerly VP of Research at NetSPI — a founding team with direct pedigree in both AI security research and traditional offensive security consulting. That background is the company’s core differentiator: rather than a static scanning or guardrail product, Dreadnode frames AI security as an ongoing offense/defense loop, using automated AI agents to continuously probe other AI systems for exploitable weaknesses.

Dreadnode raised a $14 million Series A (reported in 2025, with some date disagreement across outlets) from investors including Decibel, In-Q-Tel, Sands Capital, Next Frontier Capital and Indie VC. In-Q-Tel’s participation signals interest from the U.S. intelligence community in offensive AI testing capability, though Dreadnode does not publicly name specific government or enterprise customers.

Innovation Matrix Assessment

Innovation Velocity 7/10

Three distinct products (Crucible, Strikes, Spyglass) shipped within a few years by a small, research-heavy team; founders bring direct offensive-AI research pedigree from NVIDIA and NetSPI.

Operational Value 6/10

Gives security teams a dedicated way to continuously test AI/agentic systems pre- and post-deployment, but Spyglass remains in limited early access, so operational reach today is narrow.

Market Momentum 5/10

A real $14M Series A from credible investors including In-Q-Tel is independently verifiable, but no named enterprise or government customers are public and the flagship production product is still early-access.

Category Disruption 5/10

Offensive AI red-teaming is a meaningful niche but already contested by Mindgard, HiddenLayer, Cisco AI Defense (ex-Robust Intelligence), and CalypsoAI; Dreadnode is a strong entrant, not yet a category-definer.

Real-World Efficacy 4/10

No named incident response, independent benchmark, or published customer case study was found; efficacy claims are currently vendor-asserted only.

Enduring Relevance 7/10

Demand for adversarial testing of agentic AI systems is structurally growing, and government/IC investor interest (In-Q-Tel) suggests durable institutional relevance over a 3-5 year horizon.

Why CISOs Should Care

Gives security teams a way to continuously red-team AI agents and LLM applications using the same automated adversarial techniques real attackers would use, rather than one-time manual pentests.

What Makes It Different

Founding team combines NVIDIA's internal AI red-team leadership with NetSPI's offensive-security research background, and the product treats AI security as an ongoing agent-vs-agent contest rather than a static scan.

The Matrix Verdict

57/100 — INCREMENTAL INNOVATOR

Incremental tier: a credible, well-funded, well-pedigreed entrant in offensive AI security, but with a still-maturing production product, no public customer evidence, and strong existing competition — it has not yet shown the adoption or disruption evidence to rank higher.

Editorial Note: Claims vs. Verified Findings

The $14M Series A and investor list (Decibel, In-Q-Tel, Sands Capital) are independently reported by SecurityWeek and trade press. Claims about being 'trusted by enterprises, government agencies and frontier AI labs' are vendor-sourced and unverified against named customers.

Sources