AI CONTROL + SCALABLE OVERSIGHT
AI Control Requires a Governed Context Layer
Operational Context Transfer Protocol
A permissioned, auditable control plane for persistent AI systems.
Persistent AI agents increasingly operate across sessions, models, tools, and organizations. Yet the state guiding their behavior—goals, instructions, permissions, evidence, tool state, and prior decisions—is still commonly carried through transcript replay, opaque memory features, or proprietary orchestration.
That state is not merely memory.
It is an attack surface, a governance boundary, and an evidentiary record.
Memory asks what an AI system can recall. Control asks what it is authorized to carry forward, where that context came from, whether it changed, and whether the destination should accept it.
OCTP treats operational context as a governed object
The Operational Context Transfer Protocol defines a controlled lifecycle for transferring operational state:
- 01Select
- 02Authorize
- 03Seal
- 04Transfer
- 05Validate
- 06Accept or Reject
- 07Audit
A conforming system can support:
- explicit authority and scope;
- provenance and integrity verification;
- bounded, sealed transfer packages;
- destination-side validation and gating;
- human review and acceptance;
- discrepancy and recovery records;
- auditable custody across models and vendors.
The objective is not maximum memory. It is controlled continuity: enough retained state for useful, long-lived agents, with boundaries and evidence strong enough for meaningful oversight.
What exists today
OCTP is not merely a proposed research direction.
A patent application is pending, and a working reference environment—GM Library—has been built to test the architecture in practice.
Current implementation work includes:
- bounded context packages with cryptographic manifests;
- authority, consent, refusal, and qualification records;
- separated source evidence and interpretation;
- explicit acceptance and freeze boundaries;
- destination validation and recognition testing;
- discrepancy, rupture, and recovery documentation;
- multi-model transfer and verification exercises;
- durable local custody with independent backup and restoration procedures.
GM Library functions as a proof-governed reference implementation and proving ground. It is not presented as a finished enterprise product.
The company opportunity
A dedicated company can turn this architecture into neutral infrastructure for AI control and scalable oversight.
The product layer could include:
- an OCTP SDK and API;
- policy-driven context packaging;
- authorization and scope enforcement;
- transfer-package signing and validation;
- vendor-neutral custody and audit records;
- human approval and exception workflows;
- model and agent integrations;
- incident reconstruction and independent verification tooling.
Potential applications include persistent enterprise agents, regulated AI workflows, cross-model migration, multi-agent systems, high-consequence assistants, independent evaluations, and investigations of unexpected agent behavior.
Before an agent is allowed to carry operational state into a new session, model, tool, or environment, an organization should be able to determine exactly what is being transferred, who authorized it, whether it was altered, and whether it should be accepted.
Why now
Models are becoming persistent actors before a credible governance layer for persistent state exists.
Current oversight methods concentrate on model outputs and tool calls. But the context that shapes those outputs and calls may be hidden, stale, corrupted, overbroad, or inherited without clear authority.
Oversight cannot scale when the operative state itself is invisible.
OCTP makes that state inspectable, permissioned, portable, and accountable.
Founder
Kerry Bryson is a practicing attorney with a prior background in systems administration and network engineering. He developed OCTP after years of practical work with context continuity, model transitions, multi-model workflows, and the failure modes of opaque AI memory systems.
He filed patent protection around the protocol and built GM Library as its working reference environment.
The opportunity is to form a dedicated company around OCTP-compatible control, verification, and oversight infrastructure while GlitchMob.ai remains the originating research and reference-implementation environment.