Washington, DC
RISU studies what must remain true as machine-readable meaning crosses evidence, interfaces, and action boundaries. The work turns those distinctions into bounded definitions, executable checks, and inspectable public records.
Washington, DC
Research news
Recent work spans agent security, software supply chains, public-sector AI, and access to AI infrastructure.
Open conference website
Global Access, Unequal Control: The Locality Tax in Managed AI
Research on which countries have documented AI processing choices, how those choices change with model and purchase terms, and what public buyers should verify.
Open official conference program
Corrective Action in Medicaid Appeals: Legal Authority and Source Retrieval
Traces the governing regulations and contract versions in Medicaid appeals to identify legally supported corrective actions.
Open conference website
When One Tool Means Many Actions: A Conformance Criterion for Policy-Preserving Agent Interfaces
Develops a criterion for determining whether agent-facing interfaces preserve the safeguards of the operations they represent.
Open conference website
When Trial Data Portals Move: Cross-Sponsor Policy Mismatches
An archival audit identifies three cases where a clinical-trial data portal paired a sponsor's transition notice with another sponsor's policy text, supported by publicly reproducible historical web records.
Open workshop website
What Must a Web-Agent Log Remember? Candidate Omission, Evidence Collision, and Selective Retention
Investigates which candidate actions, conflicting evidence, and retention choices an agent log must preserve for reliable later review.
Open conference website
Runtime Reachability of Security Repairs in PyPI
Measures whether published Python security fixes remain reachable on existing runtimes, using historical PyPI advisories and release metadata.
Image credits: UTE campus, Elqueeditayredacta (CC BY-SA 4.0; cropped to fit) · Phoenix skyline, J550112 (CC0).
Tests whether agent-facing interfaces preserve the behavior and safeguards of source operations.
Investigates which consequences are determined, and which remain unresolved, when a machine takes action.
Studies when a system can rely on verified evidence before a wider workflow has finished.
Executable checks for consequence preservation across agent-facing tools, with inspectable evidence and browser-verifiable handoff records.
