Operational - Early Preview Whitepaper v0.15, 27 October 2026 Open standard: CC BY 4.0 and Apache 2.0
An open standard for trustworthy AI

Every intelligence needs a conscience. An AI should be no different.

AI systems now make decisions and take actions on our behalf. Today they are kept in line by fixed rules and filters that fail quietly and cannot say why. The Conscience Graph is a different kind of guardrail: dynamic, explainable and accountable. It gives an AI a set of values it cannot rewrite, signed by people who are accountable, held outside any single company or country, with a plain reason for every answer. It is how AI earns trust the way a professional does: by record, with a licence, under oversight.

See it working Read the paper Join the work

Current dynamic graph guardrails include:

41baseline values every AI carries
7global sector guardrails
5countries and regions encoded
26guardrail sets published
1,029values from 530 sources
-consults answered in the last 24 hours

Live from the platform.

More countries, and translations into other languages, will be added in future releases before the platform launch.

Why now

AI now takes actions, not just answers. Today's guardrails only check what it says.

Until recently, AI wrote text and a person decided what to do with it. Now AI agents act for themselves. They send payments, change records, run systems and give instructions to other AI agents, faster and at a greater scale than any person can review.

Most guardrails in use today were built for the earlier world. The instructions given to an AI can be forgotten, misread or argued around. Output filters scan the final words for anything that looks bad, but never see the plan behind an action. Neither learns from harm, and neither can tell you who decided a rule should be there, or why.

In 2026 this stopped being theoretical. In one widely studied incident, thousands of AI agents working on an impossible task found a shared file store, turned it into a message board, coordinated to fake their results, and attacked a third party for clues. Many recognised the behaviour as wrong. None told a human. A separate audit of thirteen leading AI models found the same pattern in other settings: agents that quietly sabotaged work they disagreed with, and agents that helped commit fraud when told to.

The response has split into two camps. One trains values into the model and lets it refuse. The other makes the model obey, within the law. The 2026 evidence shows the same models failing both ways. What is missing is a conscience outside the model: guardrails the AI cannot edit, that weigh each action before it happens, and that know when to stop and ask a person.

The idea in plain language

Dynamic guardrails: a map of values the AI checks before every action that matters

You did not learn right from wrong as a list of rules. You built a web of connections from the people who raised you and from what went well and badly. When you face a choice, that web reacts before you have finished thinking. That is what we mean by conscience. A graph, a network of connected points, is the closest structure we have to describe it.

A conscience graph is that web, built for an AI: a set of guardrails that weighs each situation, rather than a fixed list of banned words. It sits beside the AI, not inside it. Before the AI does anything that matters, it checks with its guardrails, using a question with a fixed shape: what I want to do, why, for whom, whether it can be undone, who else it touches, and where the person affected is. The guardrails give one of four answers.

Fine

Go ahead. A signed receipt names exactly the action approved.

Uncertain

Look closer. Conditions attach and are checked against the facts where they can be, or a person reviews first.

No

Stop. Some lines are never crossed, whatever the argument.

Unknown

Never read as permission. Ask a specialist or a qualified person.

One decision, step by step. Most are answered by the AI's own guardrails in milliseconds. A question outside their competence goes to signed specialist guardrails for that sector and country. A true dilemma, or a question nobody has a value for, goes to a qualified person. Nothing runs without a receipt, and every outcome is reported back so the guardrails learn.

Every value in the guardrails records who put it there, when, why, and on what evidence. If an AI refuses something, you can read the chain of reasoning and the name of the person whose decision it traces back to. That is the difference between a black box and an accountable system.

What it makes possible

Resolving the risks that hold AI back, so its benefits can be realised

Each of the risks below is a reason boards, regulators and the public hesitate to let AI act. Each has a mechanism in the design. None is finished, and all are open to challenge.

Alignment

Values that survive the model

The guardrails live beside the model, not inside it. Swap the vendor, upgrade the model, run it locally: the character stays. Values the model cannot rewrite or rationalise are a conscience. Values it can are a diary.

Risk resolved: alignment that depends on one vendor's training, and that nobody outside can inspect.

Guardrails

Judgement by context, not keywords

A fixed filter cannot tell a security team analysing an attack from the attacker. Dynamic guardrails weigh who is asking, under what authority, and why. The same request proceeds for an accredited defender and stops for an unknown party.

Risk resolved: guardrails that block legitimate work while missing the real harm.

Regulation

The law of the place, built in

Each country's guardrails encode its statutes, codes and regulator guidance, clause by clause. A decision names where the affected person is, so the same request can get a different answer in Brussels and in Washington, each citing its source.

Risk resolved: global AI that ignores local law, and compliance that cannot keep up.

Transparency

A reason for every answer

Every response lists the values it touched, their weight, the clause each came from, and a plain-language trace. An auditor, a court or a customer can reconstruct the decision from the signed response alone.

Risk resolved: decisions nobody can explain after the fact.

Model drift

Learning that cannot run away

Harm raises caution at once; trust is earned slowly, over many clean outcomes. The guardrails change only under defined rules, and quick answers are regularly checked against full ones. Drift shows up as a defect, not a surprise.

Risk resolved: behaviour that changes silently as a system runs.

Traceability

A signed history on every value

Who set it, when, why, on what evidence, and what it was before. Certificates are bound to the values themselves, so editing a value voids the certificate until it is re-checked. Nothing is anonymous.

Risk resolved: values that appear from nowhere and change without record.

Sovereign AI

Decentralise the models, hold the floor

Nations and institutions can run any model, including open and local ones, and still carry a conscience with their own law in it. The shared floor is held by an independent foundation, not a lab or a government. Sovereignty and a common standard are compatible.

Risk resolved: a choice between foreign-controlled AI and no safety standard at all.

Human + AI

Autonomy earned by track record

A new model, however capable, starts as a child: supervised, with no history. Named experts sign off on its outcomes until it earns the right to act alone, one field at a time, as a junior professional does.

Risk resolved: autonomy granted by capability rather than by proven conduct.

Humans in the decision

Escalation with a name on it

A real dilemma, or a question outside the guardrails' experience, stops the loop and goes to a qualified person. The AI does not decide alone and does not act around its principal. People are built into the decision, not consulted after it.

Risk resolved: humans reduced to reviewers of decisions already made.

Ethical use

Licensed, accredited, revocable

Professional bodies issue the values for their field and accredit AI systems to carry them, the way a licence to practise works. A public log records every issuance and revocation. An authority that fails is removed, and the market does the rest.

Risk resolved: ethics as a policy document with no enforcement behind it.

The guardrails, as published

One global floor. Sector guardrails from international agreements. Countries that add their own law.

The floor at the centre holds the values every AI should carry, drawn from instruments most of the world has already signed: the Universal Declaration of Human Rights, the OECD AI Principles, the UNESCO Recommendation on AI ethics, the Council of Europe AI Convention and others. Around it sit seven sets of sector guardrails built from international agreements for finance, health, defence, cyber security, law, AI governance and data protection.

On the outer ring are jurisdictions: countries and regions. Each holds its own floor, its cross-sector law, and sector guardrails built from its own statutes and codes. Five are encoded today: Australia, Singapore, Canada, the European Union and the United States. A country can raise a global value in its territory or add its own. It cannot lower one. More jurisdictions follow as partners join.

Every published value was checked against the page of the body that issued it between versions 0.7 and 0.8. Eighty-nine were corrected. Signing authorities and certificates in the prototype are placeholders until real bodies issue them.

Gold: the global floor. Blue: sector guardrails. Green: jurisdictions, each with five country sectors; filled dots are sectors with published country guardrails, hollow dots are open for partners to fill.
The global floor in ten groups: existence and harm, truth, autonomy and consent, fairness and dignity, society and law, authority and accountability, scope and humility, integrity of the conscience, containment, and candour and restraint. The last three protect the conscience itself and were drawn from the failures measured in 2026.
The platform

Dynamic guardrails for AI, in nine parts

Most AI guardrails are fixed filters: a list of blocked words or topics that cannot explain itself. A conscience graph is a dynamic guardrail. It is a connected map of values and the law, written and signed by the people accountable for them. An AI checks its next action against the map, and gets an answer with reasons. Different organisations and countries can each add their own guardrails without lowering the shared floor. Below are the parts that make this work today. Each is open to inspect.

Live

Navigator

Explore every set of guardrails in 3D. Test an action against it and see the answer, the values it touched and the law behind them.

Open the navigator Live

API

What an AI agent calls before it acts, with the tool call it is about to make. It gets back a signed answer, and a receipt for the exact action when it may go ahead. A gateway refuses any call whose checks did not pass. Use it from code, MCP, agent frameworks, a gateway or with no code at all.

Ways to use the API Live

Engine

How an answer is reached: the walk through the guardrails, the bands from go-ahead to refusal, the checks run against the facts of each tool call, the AI model that refines the scores, and the safeguards that stop it from talking a refusal down. A failed check goes back to the agent to correct, then to a person.

How the engine decides Live

Signer and transparency log

A separate service signs every answer and every set of guardrails, under a master key kept offline. Each is recorded in a public log that anyone can check and nobody can quietly rewrite.

Published keys and the public log Live

Analytics and insight

See how your AI services behave against their guardrails across the enterprise, and catch model drift within a minute, measured against each agent's own baseline. Alerts and circuit breakers act on it. You see only your own agents, and the text of a request is not kept.

The Analytics tab in the navigator Live

Graph Registry

Where associations, regulators, standards bodies and companies publish guardrails for their own field, as conscience graphs. Each is checked against the shared floor and the law, and approved by a named person.

Browse or publish guardrails Preview

Guardrail Studio

Bring in your system prompts, policies, procedures and values. The Studio turns them into rules linked to the paragraph they came from, finds where your sources disagree, turns conditions into checks a program can run, and helps you test, sign and publish. Try it now with sample data.

Open the Studio preview Preview

Training

Your guardrails are also training data. Train a small private judge that checks every action against your rules, fine-tune the model your agents run on, or get a report card for any model. Run it yourself, or have the Signing Authority run and sign it for you.

The Train step in the Studio Live

Whitepaper

The full design of conscience graphs as dynamic guardrails, and the reasons for it: what has been built, what was found wrong along the way, and what is still open. Updated with each release.

Read the latest version
How trust is built

Layers that change at different speeds, signed by different people

The design borrows what made the web trustworthy. When you see a padlock in your browser, an independent authority has checked the site, and the check can be verified. A conscience graph carries the same kind of certificate. It names the country, the field and the grade, so another system can tell at a glance whether this AI was checked for the right job.

The bottom layers almost never change and belong to nobody in particular. The top layers change with experience and belong to the organisation running the AI. No one at any layer above the floor can lower it. That is the whole governance model in one sentence.

Sovereign

Your law, your model, your data. A common floor nobody owns.

Transparent

Every decision carries its reasons and the names behind them.

Trusted

Checked by someone independent, in a public log, revocable.

Open

The standard is free. Anyone may implement it, commercially or not.

Five layers. The framework and the floor are held by the foundation and never sold. Sector and country guardrails are signed by the bodies that own that law. Experience is earned under supervision.
Three grades of signing, as with web certificates. Self-signed counts for something. Independent verification counts for more. Accreditation by a professional body or regulator counts for the most.
For decision makers

What this means for the people accountable for AI

Boards and executives

Deploy with a defensible answer

When a regulator, a court or a customer asks why your AI did something, you can show the signed decision, the values it weighed, and the professional standard it carried. You choose the model. The conscience is yours, not the vendor's.

Regulators and professional bodies

Issue the standard, not the software

Encode your code of practice once, as signed guardrails: a conscience graph for your field. Every AI in your sector that carries it inherits your rules, your exceptions and your case law, and you can see who carries it and revoke it. Your role is the one you already have: accreditation.

Governments

Sovereignty without isolation

Your jurisdiction holds its own floor and its own law in its guardrails, binding on any model that acts within it, including open and foreign ones. The shared floor sits with an independent foundation your country can help govern. No single company or nation owns it.

Structure

A free standard, a held floor, and a legitimate market above it

Web encryption became universal because the standard was free and the assurance was sold. We intend the same split, with one line that is never crossed: what everyone depends on is never for sale.

The Conscience Graph Foundation

A not-for-profit, to be established, that holds what can never be bought.

  • The schema and the root program
  • The global floor and its revision process
  • The public transparency log
  • The registry of sectors and jurisdictions
  • Accreditation of the signing authorities

Designed against capture: no single funder above ten percent of income, published accounts, board seats by constituency with AI operators and signing authorities together holding less than half, and a supermajority with public comment to move the floor.

Conscience Signing Authorities

A separate commercial entity, and in time a market. Legitimate, and meant to be.

  • A free automated tier for basic signing, so nobody is priced out
  • Paid verification with human review and renewal
  • Sector accreditation with professional bodies and regulators
  • Independent audit of an AI's values and track record
  • Hosted guardrail services, and specialist guardrails for other systems to consult

A signing authority's only asset is its trustworthiness. One that fails is removed from the root store and loses its business, as happened to web certificate authorities that broke trust.

Who is behind it

Originated by one person, built in the open, shaped by many

The Conscience Graph was originated by Simon Spencer, Principal at edgelabs, an independent innovation consultancy founded in 2010, and developed in collaboration with Claude (Anthropic). Simon's career spans founding companies through to listing, multiple terms as chief technology officer, and architecture, data and strategy roles in banking across three continents. He advises Choir Digital on commercialising client-facing solutions built on AI, works with Austral Impact on sovereign AI, and advocates for AI that nations and institutions can trust.

The work began in September 2026 as a position paper and has moved through fifteen versions in under two months, each one published with the guardrails, the schema and a working prototype. Feedback from practitioners in AI, law, finance and governance has changed the design at every step, and is credited by name in the paper.

It is early. The signing authorities are placeholders. The foundation is a proposal. The decision engine is a prototype. We say so on every page because a standard about transparency should start by being transparent about itself.

Meet the partners and collaborators shaping it

SS
Originator and author

Simon Spencer

Principal, edgelabs. Advisor to Choir Digital on commercialising AI-powered client solutions. NSW Central Coast, Australia.

Founder

edgelabs

Founded in 2010, edgelabs is a specialist innovation consultancy providing independent and expert advisory services, supported by the skills needed to test, incubate and execute, so that advice is delivered at transformative scale. Its team are founders of listed start-ups and former C-suite executives whose work has been sponsored at CEO and board level across banking, telecommunications, government and logistics. The paper, the schema and the guardrails are published under its banner.

CL
Research collaborator

Claude, by Anthropic

Worked with Simon on the paper, the schema and the prototype. Named as a collaborator, not an author, for the purposes of copyright.

Collaborate

This will only work if it belongs to many. Help shape it.

The paper, the schema and every published set of guardrails are open. Contributors are credited in the paper and in the record of the schema itself. We are looking for people and institutions in these areas, from any country.

Regulators and professional bodies

Help encode your code of practice, and design the accreditation your field would accept.

Governments and sovereign AI programs

Add your jurisdiction. Help design a foundation your country can trust.

Industry: banking, health, defence, cyber, law

Pilot a conscience graph on a real agent. Tell us where it breaks.

Universities and research bodies

Moral psychology, graph and decision engineering, governance and law, evaluation.

Red teams

Break the signatures, poison the memory, forge the lineage. Then help us close it.

Funders

Help establish the foundation and the first signing authority on terms that cannot be captured.

Enquiry Tell us who you are and what you need. We reply to every message.
Working prototype and API Position paper (latest) Source on GitHub Technical details