Research
Authorization-Preserving Context Lifecycles for Long-Horizon AI Agents
Authorization boundaries throughout long-horizon agent context lifecycles.
Authorization boundaries throughout long-horizon agent context lifecycles.
Authority conservation as agent populations replace, fork, and roll back software.
Budget conservation and fault tolerance in distributed multi-agent delegation.
A paired cross-lingual study protocol for token use and reasoning efficiency.
Runtime authorization of subgoals generated by long-horizon tool-using agents.
A root-scoped revocation protocol that fences delegated and asynchronous old-root paths while preserving work backed by current, independently sufficient authority.
A provenance-bounded runtime that quarantines agent-acquired resources, resolves their real capabilities, and activates only authority within a current relational policy envelope.
A reference monitor that binds authorization to the exact transitive dependency closure of an agent skill and revalidates that closure when a durable effect is produced.
A reproducible measure of migration-associated occupied-dwelling use, evaluated for incremental forecasting value against Vancouver new-house price benchmarks.
A claim-carrying mediation layer that converts decision-relevant rationale content into typed action claims and verifies them against current server-held facts before governed execution.
A server-side authorization layer that binds tool visibility and proposed effects to the user’s current expressed intent—narrowing, never expanding, existing authority.
A governance-first specification for exposing application tools to AI agents through server-side authorization, reviewable writes, predictable failure semantics, and audit.