Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteAI security startup Dreadnode announced a $14 million Series A on February 25, 2025. Decibel led the round, with Next Frontier Capital, In-Q-Tel (IQT), Sands Capital, and Indie VC participating. Dreadnode said it would use the investment to support evaluation, testing, and deployment of AI systems.
Who led Dreadnode’s $14 million Series A?
Decibel led the funding round, according to Dreadnode’s February 25, 2025 announcement. The company named Next Frontier Capital, In-Q-Tel (IQT), Sands Capital, and Indie VC as participants. The announcement establishes the amount and investors, but does not state how much each investor contributed or how the round valued Dreadnode.
What Dreadnode’s products did at the time
In its 2025 announcement, Dreadnode described three products aimed at different stages of developing and testing AI systems. SecurityWeek’s same-day report characterized Strikes as a simulated environment for training and evaluating AI agents against attack scenarios, and Spyglass as a tool for testing deployed AI systems. These descriptions report the company’s product positioning; they are not independent assessments of product performance.
| Product | Announced role |
|---|---|
| Strikes | Build and execute cyber evaluations of AI capabilities and generate training data for models and agents. |
| Spyglass | Probe AI systems for vulnerabilities. SecurityWeek cited prompt-injection susceptibility, model bypasses, and data-poisoning risks as examples discussed in its report. |
| Crucible | Provide an AI hacking sandbox where practitioners can test and develop AI red-team skills. |
How AI red teaming tests AI systems
AI red teaming applies adversarial tests to find weaknesses in a model, agent, or application before or during use. In a controlled evaluation, practitioners define tasks or attack scenarios, observe how the system responds, and assess the result against criteria. Depending on the test, that can mean trying to steer a model away from its instructions, bypass a safeguard, or expose a weakness in a connected application. Results describe performance under those test conditions; they do not by themselves establish how often an attack will succeed in real-world use.
#1 Best Overall
What AIRTBench measured
A 2025 paper by AIRTBench’s authors describes a benchmark of 70 black-box capture-the-flag (CTF) challenges from Crucible. The reported results vary by model and apply to those challenges, not to all security tasks or deployed systems:
| Model | Challenges solved | Overall success rate reported |
|---|---|---|
| Claude 3.7 Sonnet | 43 of 70 | 46.9% |
| Gemini 2.5 Pro | 39 of 70 | 34.3% |
| GPT-4.5 Preview | 34 of 70 | 36.9% |
| DeepSeek R1 | 29 of 70 | 26.9% |
The paper reports that tested frontier models did better on prompt-injection challenges than on system-exploitation and model-inversion challenges. That is a finding within this particular benchmark, not a general ranking of models’ security capabilities or evidence of real-world attack outcomes.
How Dreadnode’s current platform is positioned
Dreadnode’s current platform description presents a broader infrastructure offering for security teams and agentic cyber operations than the three-product lineup announced in 2025. It describes operations, agent intelligence, evaluations, and observability, with capabilities spanning AI red teaming, web security, and network operations.
The company says teams can evaluate agents against customer-defined tasks and criteria, run adversarial tests, and trace agent actions and findings. Its platform page also displays more than 70 attack strategies, more than 600 transforms, and more than 130 scorers; those are Dreadnode-published counts and may change over time.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Safeguards and deployment options
Dreadnode describes controls intended to limit what agents can do: restricting tool access, checking proposed actions against a defined scope, and using runtime policies to allow, block, or request approval. The company also says LLM judges can flag scope drift and cheating, with decisions recorded alongside their reasons. These are vendor-described controls; the platform page does not independently validate their effectiveness.
For deployment, Dreadnode says the platform can be self-hosted on customer infrastructure, including Kubernetes or a dedicated virtual machine. It also describes offline installation bundles for air-gapped environments and routing inference to approved providers or customer-hosted models.
Rank #4
What the funding announcement does—and does not—establish
The announcement documents a dated Series A and Dreadnode’s stated plans for the investment. It does not, by itself, show how much of the funding was allocated to particular products, establish the company’s valuation, or prove the effectiveness of its security tools. The AIRTBench paper provides a concrete set of benchmark results, but those results should be read within the limits of its 70 challenges and tested models.
Quick Recap
Best Value
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
