Products
Shared sign-in, permissions, audit history, and billing.
Research
Publications on AI model behavior: behavioral fingerprinting, sycophancy, adversarial resilience, and context degradation.
Publications
Government
Prism for model evaluation before adoption. Pallette on agency infrastructure. Proving Ground for isolated analysis.
Government
Deployment and security
Trust & safetyHosting
Presba cloud, customer cloud tenant, or on-premises.
Air-gapped
All products install on networks without internet access.
STIG
Built to the DoD Application Security and Development STIG.
Access control
Shows who can open an item, and why, before it is shared.
Recent publications
All publications
Behavioral evaluationBehavioral Fingerprinting of Large Language Models: Beyond Benchmark Scores to Operational CharacterizationMarch 2026
Agreement biasThe Sycophancy Problem: Measuring and Mitigating Agreement Bias in Large Language ModelsMarch 2026
Adversarial testingAdversarial Resilience Testing for Production AI Systems: A Graduated Escalation MethodologyMarch 2026
Contact
Demonstrations, pricing, and pilots.