Open-source adversarial security for the agentic AI era.

PwnKit Labs builds open-source tools that find failures before attackers, customers, or production do.

• pwnkit: Attack-driven evaluation and autonomous pentesting for web apps, source code, packages, and AI systems.
• foxguard: Linter-fast security scanning and policy checks to prevent regressions before deploy.
• opensoar: Response and orchestration for routing validated failures into real operator workflows.
• Noeris: PwnKit Labs Research OS for autonomous empirical discovery, currently led by a flagship GPU kernel optimization track.

Most AI reliability tooling measures prompts, traces, and outputs.

We care about whether a capable system can be broken, exploited, or pushed outside its intended bounds.

That means:

• adversarial testing, not just happy-path evals
• evidence-backed findings, not vibe-based scores
• prevention and response, not detection alone

AI agents are being shipped into real workflows faster than teams can verify them. The bottleneck is no longer generation. It is trust.

PwnKit Labs exists to close that gap.

The org is being consolidated around the core company products now.

• pwnkit is the flagship wedge
• pwnkit-cloud is the hosted/private runtime surface
• foxguard and opensoar are sibling products in the same company story

Noeris currently sits adjacent to the core security products: it strengthens the shared research, optimization, and evaluation substrate that future Labs products can build on.