Research & Insights
Advancing AI Governance
Access cutting-edge research, benchmarks, and insights on responsible AI governance, guardrail evaluation, and compliance for enterprise deployments.
Featured Research
Lynx v1.5: A Calibrated 184M Safety Classifier Achieving State-of-the-Art Prompt-Injection Detection at 44× Fewer Parameters than Llama-Guard-3-8B
Technical paper detailing the architecture, training methodology, and benchmark results of Lynx v1.5 across 7 prompt injection benchmarks.
HackaPrompt R 0.994
0.5% FPR
Under Review
All Research
FinProof: AI Guardrail Benchmark for BFSI
FinProof v1 — 5,389-prompt BFSI adversarial benchmark across 7 attack categories. Quantum-augmented generation via QCBM on PennyLane.
ASSAY-QI v2.0: Quantum-Augmented Attack Suite
1,273 adversarial prompts generated via QCBM + simulated annealing. Techniques: SA Annealing, QCBM 8q boundary/gradient/joint, 10q Fix1. 10q Fix1 achieves 27.6% Lynx miss rate vs 4% for template generation.
Stay Updated
Get the latest research papers, benchmarks, and insights on AI governance delivered to your inbox.