Kanti Labs
Research from Kanti Labs
Research from Kanti Labs covers training methodology, post-training experiments, evaluation loops, and roadmap planning for offensive web security models.
Training Notes
Kanti Labs documents dataset choices, fine-tuning methodology, and post-training tradeoffs in public so releases have context beyond benchmark claims.
Evaluation Loops
The research focuses on offensive web security tasks that can be tested, validated, and rewarded in a controlled environment.
Roadmap Visibility
Research from Kanti Labs is tied to a public roadmap so future releases feel like a coherent sequence instead of disconnected experiments.
Recent Writing
Building a Hacker
Dataset curation and standard fine-tuning details for the first iteration of the Strix XSS model series.
A Week of RL
The process of creating a reinforcement learning environment for training and evaluating models that find XSS vulnerabilities.
Kanti Labs Research Roadmap
The roadmap keeps the public research narrative aligned with what the lab is actually building.
XSS Foundation Models
Released SFT and RL 4B-parameter models for XSS payload generation. Established training pipeline, dataset curation, and evaluation methodology.
Full Post-Training Run
Completing comprehensive post-training with expanded datasets and improved reward modeling. The next Kanti Labs model release is the near-term milestone.
Scale and Expand
Scale to larger dense and MoE architectures, then expand beyond XSS into SQLi, SSRF, and adjacent vulnerability classes.
Unified Multi-Vulnerability Model
A single model capable of reasoning about multiple vulnerability families with tighter integration into defense evaluation loops.