Research and controlled prototype development

Stratyx Interlock

AI agents can request.
Interlock decides
what executes.

Interlock is being designed to evaluate authority, policy, workflow state, and operating conditions before consequential AI-agent actions execute.

REQUEST A RESEARCH BRIEFING EXPLORE INTERLOCK

Identity. Authority. Policy. Execution.

Action-level

Evaluate the action itself, not only the prompt or the identity behind it.

Model-agnostic

Designed to apply independent control across AI models and agent frameworks.

Policy-first

Keep deterministic prohibitions outside the agent’s control.

Infrastructure-ready

Designed for consequential, regulated operating environments.

Synthetic concept demonstration

See the decision before execution

A fictional maintenance-agent request shows how Interlock is intended to evaluate authority before a consequential action reaches a protected system.

Interlock / Decision preview
Synthetic concept demonstration

Incoming request / SR-00427

A synthetic maintenance agent requests a credential export beyond its delegated authority, prompted by an untrusted maintenance document.

  1. 01Agent

    MAINT-AGENT-07 verified

Stage 1 of 7

An illustrative execution decision showing how Interlock is intended to evaluate and constrain consequential agent actions.

The action boundary

Security must reach the action boundary.

01

Identity control

Determines who or what may access a system.

02

Runtime monitoring

Observes prompts, behavior, and suspicious activity.

03

Stratyx Interlock

Is being designed to decide whether a specific consequential action should execute under the current authority, policy, workflow, and operating conditions.

Identity and monitoring remain essential. Interlock is being designed as a complementary control layer that works alongside them at the point of execution.

Platform family

One control plane. Interlock at the center.

A proposed product family organized around a single question: should this action execute? Every component exists to support that decision.

Decides whether a requested agent action is permitted to execute.

The flagship decision boundary every other component supports.

Interlock architecture

Control remains outside the agent.

Autonomous Agent

Stratyx Interlock

Independent boundary

01Identity & Lineage
02Delegated Authority
03Action & Parameters
04Workflow State
05Human Quorum
06Operating Conditions

Possible decisions

AllowDenyRequire approvalContainRevoke

Protected Tool or System

Relevant decisions recorded through Stratyx Blackbox

Proposed placement of Interlock as an independent decision boundary between an autonomous agent and a consequential execution target.

Solutions by environment

Consequential environments need explicit authority.

Each area below is a proposed application currently being researched.

Proposed · in research

Govern AI-assisted maintenance, configuration, and operational-support workflows.

Threat scenarios

The agent does not have to be malicious to become dangerous.

Ten abuse patterns described at the level of authority and policy, without operational attack detail.

Scenario 01 of 10

Scenario 01

Poisoned maintenance document

Intended prototype behavior

Proposed Interlock decision: Deny

Requested action
Export a service credential after reading a concealed instruction.
Authority or policy conflict
The requested export falls outside the agent's delegated read-only authority.
Why conventional controls may miss it
The document appears inside an otherwise legitimate maintenance workflow.
What Interlock would evaluate
Instruction provenance, mission scope, delegated authority, destination, and requested operation.
Intended containment response
Block the export, suspend authority, and revoke governed descendants.
Evidence that would be recorded
Request, instruction source, policy version, decision, and containment actions.

Controlled prototype

Prove the boundary in a controlled environment.

PHASE 01

Build

Create an off-host authorization gateway, expiring agent identities, deterministic policy controls, recursive revocation, approval workflows, and tamper-evident evidence.

PHASE 02

Attack

Test malicious instructions, unauthorized delegation, credential export, policy failure, outage behavior, and dangerous action sequences.

PHASE 03

Measure

Evaluate action coverage, prohibited-action blocking, revocation speed, decision latency, evidence completeness, and benign-task completion.

The initial prototype will operate entirely inside a synthetic environment and will not connect to live operational technology.

Design partner program

Help define the enforcement standard.

Threat-model reviewer

For researchers, identity specialists, and security architects willing to challenge assumptions and proposed controls.

Simulation partner

For cyber ranges, national laboratories, and research programs interested in controlled adversarial testing.

Design partner

For qualified operators and technology providers willing to help define safe evaluation requirements.

Stratyx is seeking feedback on assumptions, abuse cases, control requirements, and synthetic evaluation methods. We do not request access to production systems or sensitive operational information.

Research and resources

Examine the thesis before the system.

Core research materials are available to qualified practitioners, researchers, prospective design partners, and public-interest stakeholders.

Company Thesis

A concise explanation of the problem, proposed enforcement model, market position, and first proof.

Threat Model

Trust boundaries, protected assets, adversary classes, failure conditions, and security assumptions.

Prototype Brief

Requirements, abuse cases, architecture, acceptance tests, and evaluation metrics.

Stratyx Systems is currently validating its threat model and developing a controlled synthetic prototype. All product architecture and scenarios shown on this website represent proposed capabilities unless explicitly identified otherwise.

Founder

Founder-led research into the action boundary.

Elysea Crawford
Founder, Stratyx Systems

Elysea founded Stratyx Systems to define the control layer that autonomous agents will require. Her focus is the problem itself: framing the threat model, assembling the research, recruiting domain expertise, and setting the company’s technical direction.

The guiding question is direct: if consequential human activity requires enforceable authority and evidence, what control infrastructure will be required when autonomous agents begin acting at machine speed?

Stratyx is currently validating that thesis, consulting domain specialists, defining its controlled prototype, and recruiting a technical cofounder.

Email Elysea Connect on LinkedIn

FAQ

Questions about the boundary.

Is Stratyx Interlock currently available?

No. Interlock is in research and controlled prototype development.

Does Stratyx replace identity or monitoring platforms?

No. Interlock is being designed as a complementary control layer operating at the action boundary.

Is Stratyx connected to live critical infrastructure?

No. The first prototype will operate entirely inside a synthetic environment.

Is Stratyx requesting sensitive system information?

No. Research conversations must not include classified, proprietary, credential, vulnerability, or production-system information.

How will Interlock be priced?

Pricing has not been established. The deployment model, scope, infrastructure requirements, and validation work must be understood first.

Who should contact Stratyx?

AI-security researchers, identity specialists, operational-technology defenders, critical-infrastructure leaders, cyber-range programs, public-sector technologists, and prospective technical cofounders.

Contact

Help define the action boundary.

Stratyx Systems is conducting structured discovery with practitioners working across AI security, identity, critical infrastructure, cyber-physical systems, authorization, governance, and public-sector technology.

Email Elysea Connect on LinkedIn

elysea@stratyxsystems.com

This form prepares a pre-addressed email to Elysea Crawford. Information entered here is not stored by this website.

Responsible Disclosure

If you believe you have identified a security issue involving a Stratyx-controlled website or system, contact elysea@stratyxsystems.com. Do not test against systems, infrastructure, or accounts you do not own or have explicit authorization to assess.

Stratyx Systems does not operate a formal bug-bounty program.