Built to be trusted,
not just capable.
Kory is KYRO'S AI Agent — engineered so your data stays yours, your actions stay supervised, and every decision leaves a traceable record. Kory's architecture aligns with the NIST AI Risk Management Framework, NIST SP 800-53, and the NIST Privacy Framework — the standards defining responsible AI adoption across US federal and enterprise environments.
Every layer of the stack, secured.
PII never reaches the model
Personal identifiers — emails, phone numbers, names — are replaced with reversible tokens before any data enters the LLM. Real values are restored only in the trusted output path.
Delegated per-org identity
Each organization gets a dedicated Kory agent identity, auto-provisioned with uniqueness constraints. Cross-org delegation is explicitly blocked.
Runtime permission gating
Tool availability is filtered by user RBAC before the planner sees them, then revalidated again at execution. No tool is invoked without permission.
Human-in-the-loop approvals
Sensitive actions pause for explicit confirmation. Rejection cancels silently. Session ownership is validated to prevent cross-user takeover.
Full on-behalf audit trail
Every mutable action records both agent and human principal. Approval states and timestamps are written to an immutable trace.
Managed identity infra
Azure UAMI-based RBAC governs service access. Role naming is deterministic and auditable — no static secrets or shared credentials.
How controls stack in every request.
PII redaction
Input scanned; real PII replaced with tokens before processing.
Identity check
Agent identity established. User context preserved for attribution.
RBAC filter
Tool catalog filtered by user RBAC. Only permitted tools visible.
HITL gate
Sensitive tools pause. User confirms before any write occurs.
Audit record
Action runs under scoped token. Dual-principal record written.
Restoration
Tokens resolved to real values only in the final user response.
Built on the frameworks that define responsible AI.
AI Risk Management Framework
NIST's AI RMF defines how organizations should govern, map, measure, and manage AI risk. Kory's HITL approval gates and per-org scoping implement the GOVERN and MANAGE functions.
Security & Privacy Controls
Kory's RBAC gating, managed identity, and audit logs are direct implementations of the AC, IA, AU, and SC control families defined in this federal standard.
Privacy Risk Management
Kory's privacy-by-design model — where PII is tokenized at the input layer — directly enacts the Identify and Protect functions of the NIST Privacy Framework.
Safe AI & Cyber Hygiene
US Executive Order 14110 requires AI oversight. Kory's HITL architecture aligns with these directives, while CSF 2.0 governs credential hygiene.
What we guarantee, not just what we built.
Data is never exposed
PII tokenization is architectural, not a setting. Personal data cannot reach the LLM layer. Aligned with NIST Privacy PR.DS-P.
Mandatory approvals
Kory cannot act without human confirmation. The HITL gate is mandatory and rejection-safe. Aligns with NIST AI RMF GOVERN 6.1.
Full accountability
Dual-principal records carry both agent and user identity. Traceable and auditable. Aligned with NIST SP 800-53 AU-2.
The beliefs that shaped every decision.
Privacy by design
PII isolation is architectural — enforced at graph level, not configurable.
Least privilege
No tool or scope is granted beyond what the action requires.
Human oversight
Sensitive actions require explicit approval — Kory proposes, humans decide.
Defense in depth
Controls stack: permission check at planning and again at execution.
Auditability
Every action traceable to a human principal with a durable record.
Zero standing access
Infrastructure uses managed identity with deterministic RBAC.