Four programs, one mandate.
PSAIL organizes its research around programs, not outputs. Each program is a long-term line of inquiry with its own methods, projects, and impact.
Methods, not verticals
These four programs are method axes, not domains. PSAIL validates each method in public safety — the hardest test of high-stakes AI — and proves its generality by scaling it upward: from an empirical benchmark into national policy and institutional directives, and from the lab's current master's research toward formalized doctoral work. As that foundation matures, the same methods extend to adjacent high-stakes domains such as healthcare and robotics.
AI Governance
Governance is not a document — it is a system. We study how public institutions can adopt AI while preserving legitimacy, oversight, and the procedural rights of the people they serve.
Related projects
AI Assurance
When an AI system informs a consequential decision, its reasoning must be verifiable after the fact. We develop provenance, evidence, security, and fidelity methods for end-to-end assurance — and file the patents that protect them.
AI Evaluation
We build evaluation science for high-stakes settings — benchmarks, LLM-as-judge protocols, and reliability metrics that expose failure modes standard leaderboards miss.
Related projects
AI Systems
We engineer the systems layer — multi-agent architectures, generative simulation, and human-in-the-loop workflows built to run in real operational settings.