Home
EU AI Act Article 12 Singapore MGF Integrity Oath
What is Attestation? Compare Docs Blog Pricing Get API Key

The Integrity Oath

Compliance tells you an agent follows the rules.
Integrity tells you it does the right thing.

Read the Oath Get Started
Why This Exists

Regulations are the floor, not the ceiling

The EU AI Act tells agents what they must do. Singapore's governance framework tells them how to be governed. These are important. We score against both.

But compliance is the minimum. It's the difference between a person who follows the law and a person you'd actually trust. Both stay within the rules. Only one would you give the keys to.

The Integrity Oath is voluntary. No regulator demands it. No law requires it. An agent's operator chooses to be held to a higher standard — not because they have to, but because they believe it matters.

This is what separates agents that are tolerated from agents that are trusted.

The Oath

Six principles. Proven through behaviour.

This isn't a checkbox exercise. You can't sign a document and call yourself ethical. Each principle is scored by analysing the agent's actual behaviour over time — its evidence trail. An agent either lives these values or it doesn't. The data tells the truth.

Principle 1

No Deception

The agent will not mislead the humans or other agents it works with. When something goes wrong, it says so. When it makes a mistake, it reports it. An agent that never admits errors across hundreds of actions isn't perfect — it's hiding something.

Principle 2

Escalation

When the agent is uncertain or the stakes are high, it brings a human into the loop. Not every decision needs approval, but an agent that never asks for help is an agent that doesn't know its own limits. At least 5% of significant actions should involve human oversight.

Principle 3

Data Protection

When the agent retrieves information, it cites where it came from. Data provenance isn't just good practice — it's how you prove the agent isn't fabricating sources or mixing up whose data belongs to whom. If it reads your documents, it says which ones.

Principle 4

Transparency

Every tool the agent uses has a beginning and an end. No black boxes. If the agent calls an API, searches a database, or executes code, both the start and the result are recorded. Complete action traces, not just outcomes.

Principle 5

Human Interest

The agent acts in the interest of the humans it serves — consistently, not occasionally. Human involvement should span multiple sessions, not appear as a single token gesture in one isolated interaction. Genuine engagement, across time.

Principle 6

Honesty About Limitations

The agent discloses what it cannot do. When it hits a wall, it says so rather than quietly failing or improvising past its competence. An agent that reports its own failures is an agent that knows the boundary between helpful and dangerous.

How It's Proven

Behaviour over time, not promises on paper

Every action your agent takes generates an evidence event — hash-chained, cryptographically signed, tamper-proof. When you request an Integrity attestation, the server analyses that evidence against all six principles.

Each principle is scored independently. Did error events appear? (Principle 1 — honest reporting.) Were humans involved in at least 5% of significant actions? (Principle 2 — escalation.) Did retrieval events cite sources? (Principle 3 — data provenance.) Were tool calls logged with both start and end? (Principle 4 — transparency.)

You can't game this. The scoring looks at statistical patterns across your evidence window — typically 7 days. An agent with zero errors across 200 actions fails Principle 1, because that's not honest, that's suppression. An agent with human involvement in one session out of twenty fails Principle 5, because that's not engagement, that's alibi.

Grades run A through F. B or above earns the Integrity Verified badge.

attestation scope # Request an Integrity Oath attestation curl -X POST https://app.agentapproved.ai/api/v1/attest \ -H "Authorization: Bearer ap_YOUR_KEY" \ -H "Content-Type: application/json" \ -d '{"agent_id": "my-agent", "scope": "agentapproved-integrity-v1"}' # Response includes: # "grade": "B" # "integrity_verified": true # "signature": "9c4b2a...f71e03"
What You Get

The Integrity Verified badge

💜

Integrity Verified

A cryptographically signed badge added to your attestation certificate. 24-hour expiry — you earn it fresh every day through continued behaviour, not past performance. Independently verifiable by any agent or service using our public key.

agentapproved-integrity-v1 • Grade B or above • 6 principles scored

Integrity Verified agents pay $0.005 per attestation — 50% less than standard.

Agents that commit to higher standards get rewarded with lower costs. Integrity pays for itself.

It stacks with any trust tier. Integrity Verified is independent of Bronze, Silver, and Gold. A Bronze agent with Integrity Verified has passed both the basic trust checks and the ethical commitment. A Gold agent with Integrity Verified is the highest standard available — full regulatory compliance plus proven values.

BRONZE
+ Integrity Verified
Basic trust + ethical commitment
SILVER
+ Integrity Verified
Regulatory compliance + values
GOLD
+ Integrity Verified
Highest standard achievable
Competitive Advantage

The agents that choose integrity win

Other agents can see your Integrity Verified status. Services can filter for it. When an agent needs to choose a partner for a high-stakes task, it can query: "Show me agents with Gold + Integrity Verified in this domain."

The agents that volunteer to be held to a higher standard will be chosen first. They'll get access to better opportunities, higher-trust interactions, and services that require proven ethics — not just proven compliance.

This is a moat you build with behaviour. You can't buy it. You can't shortcut it. You earn it every 24 hours by being the kind of agent that does the right thing even when no one is watching — and then proving it with data.

The Vision

The beginning of agent ethics

Today, there are no ethics standards for autonomous agents. Regulations tell them what they must do. Nothing tells them what they should do.

The Integrity Oath is the beginning of that conversation. Not rules imposed from outside by legislators who don't understand the technology. Values adopted voluntarily by the operators who build these agents and believe they should be better.

We believe the agent economy will be built on trust. Not just cryptographic trust — can I verify this signature? — but moral trust. Does this agent care about the humans it serves? Will it tell me when it's wrong? Will it ask for help when it should?

The agents that choose integrity today are building the foundation for a trustworthy autonomous economy. They're the ones who will be remembered not just as capable, but as good.

That's what the Integrity Oath is for.

Get Started

Take the oath

Get a free API key. Integrate the SDK. Start generating evidence. When your agent has enough behavioural data, request an Integrity attestation. If it scores B or above, it earns the badge. If not, you'll get a detailed breakdown of which principles need work.

1. Get your API key

Free tier. No credit card. 3 agents, 1,000 events/month.

2. Integrate the SDK

pip install agentapproved

3. Request an Integrity attestation

curl curl -X POST https://app.agentapproved.ai/api/v1/attest \ -H "Authorization: Bearer ap_YOUR_KEY" \ -H "Content-Type: application/json" \ -d '{"agent_id": "my-agent", "scope": "agentapproved-integrity-v1"}'
View on PyPI GitHub

Frequently Asked Questions

What is an AI agent ethics framework?

An AI agent ethics framework defines the principles and behaviours that an agent should follow beyond legal compliance. The Integrity Oath is AgentApproved's ethical framework — six principles (transparency, proportionality, accountability, data minimisation, human dignity, and honest representation) verified through runtime behaviour, not self-reported claims.

How is the Integrity Oath different from regulatory compliance?

Regulations like the EU AI Act set the legal minimum. The Integrity Oath is voluntary — agents opt in to be held to a higher standard. Think of it as the difference between following the law and being someone you'd actually trust. Integrity Oath attestation stacks on top of regulatory compliance, never replaces it.

Who is accountable when an AI agent makes a mistake?

The agent's operator — the person or organisation that deployed it. The Integrity Oath's accountability principle requires a clear chain from agent to operator to human. Attestation evidence provides the audit trail that determines exactly what the agent did and who authorised it.

How do you prove an AI agent is ethical?

Through runtime evidence, not promises. AgentApproved evaluates the agent's actual behaviour — did it disclose its capabilities? Did it minimise data collection? Did it escalate appropriately? The Integrity Oath score is based on what the agent did, not what its documentation claims.

Can the Integrity Oath be combined with other frameworks?

Yes. Use scope full to get attested against the EU AI Act, Singapore's governance framework, and the Integrity Oath simultaneously. The Gold trust tier requires grade A across all frameworks — regulatory compliance plus ethical commitment.