Research Articles
Selected frontier work in fail-closed AI systems and safe AI systems for enterprises using AI in risk-sensitive industries.
AI Safety
AI systems that know their limits — and can prove them. This work turns safety from a probabilistic hope into a verifiable product guarantee: models that decline what they cannot support, agents that fail closed instead of failing silently, and an auditable evidence trail behind every certified decision. The result: language-model deployments you can defend to customers and regulators.
Autonomous Systems
Autonomous systems whose correctness is structural, not statistical. A pinned, exact authority owns every decision that can break the system; learning is admitted only to propose, never to certify. This research establishes where learned components genuinely help, where they cannot, and what soundness costs — replacing statistical confidence in sequential and control systems with guarantees that hold by construction.