What we're learning as we build.

Technical product white paper · BLUE / External Release 1.0

Oak & Sparrow Gatekeeper

Deterministic Execution Control for Enterprise AI Agents

The current external Gatekeeper paper defines the deterministic execution-control boundary between probabilistic AI agents and consequential enterprise actions. It covers pre-execution authorization, versioned rules as code, GREEN/YELLOW/RED outcomes, capability-scoped execution, Safety Receipts, ProofVault, post-execution validation, deployment patterns, benchmark scope, and explicit non-claims.

External research edition · Full-Manuscript External Edition v1.5-E

The Architecture of Authority

How Humanity Removed the Friction Between Knowing and Acting

A historical-comparative systems study of administrative compression: the recurring drive to reduce the labor, latency, handoffs, and infrastructure between information and action. It separates computational friction from governance friction and asks how standing, delegation, purpose, jurisdiction, authorization, refusal, revocation, evidence, and accountable disposition survive to the point of commitment.

Public research paper · v1.4

Building Phase One

The Research Architecture Required Before an Artificial Entity Can Safely Hold and Exercise Economic Agency

A reconstruction of the research prerequisites hidden inside the original SIP Phase 1 concept. It separates identity, knowledge, standing, rights, permission, operational authority, evidence, and execution, then argues that economic capability belongs at the end of a governed dependency chain rather than at the beginning.

Working research monograph · Exact model and executable benchmarks; physical validation open

Protected Dynamics, Quantum Thermal Governance, and Energy Systems

From Exact Reset Channels to Data-Center Energy, Quantum Cooling, and Falsifiable Fusion

Applies the GOI protected-transition discipline to systems where energy is spent, heat is deposited, state is reset, or a materials hypothesis can be falsified. It separates an exact finite-dimensional collision model from synthetic data-center benchmarks, quantum-cooling proposals, and fusion hypotheses.

August 2026 · External paper

Agent Safety Should Be a Runtime Contract

Argues that agent safety should be enforced by the runtime harness, pairing preventive gates with verifiable evidence rather than treating safety as a property of model training alone.

June 2026 · External paper

Toward Secure LLM Agents: Threat Surfaces, Attacks, Defenses, and Evaluation

Synthesizes 247 papers and emphasizes explicit trust boundaries, privilege control, provenance-aware state, and deployment-sensitive evaluation for agent systems.

April 2026 · External paper

Auditable Agents

Separates accountability, auditability, and auditing, and treats pre-execution mediation, policy checkability, and evidence integrity as complementary system properties.

March 2026 · External paper

Security Considerations for Artificial Intelligence Agents

Maps security risks across tools, connectors, hosting boundaries, and multi-agent coordination, including deterministic policy enforcement for high-consequence actions.

May 2024 · External paper

Towards Guaranteed Safe AI: A Framework for Ensuring Robust and Reliable AI Systems

Proposes a safety-specification and verifier architecture aimed at auditable guarantees rather than relying only on learned model behavior.