AI Usage Policy

1. Introduction

Ethics of AI is a certification platform, and we build it with AI tools. We think the honest way to run an ethics-focused product is to say — plainly and publicly — which AI tools we use, what we use them for, what data we give them, and where humans sit in the loop. This page is that disclosure.

This policy complements our Privacy Policy. Where the two overlap (for example, on data handling), they say the same thing.

2. How we use AI

We use AI tools in four places: building and testing the platform, authoring and reviewing certification questions, grading candidate responses to scenario-based questions, and generating imagery for marketing and editorial content. We list the vendors behind each of these uses in the next section, and the specific data boundaries in the section after that.

3. Tools and vendors

We currently use AI tools from three vendors. Each processes inputs and outputs in accordance with its own privacy and usage policy.

Anthropic — Claude Code and the Anthropic API

We use Claude Code to write, review, and test the platform’s source code, and to assist us in authoring and reviewing our certification questions. We use the Anthropic API in our grading pipeline to evaluate candidate responses to scenario-based questions. See the Anthropic Privacy Policy.

OpenAI — Codex

We use OpenAI’s Codex agent to write, review, and test the platform’s source code. See the OpenAI Privacy Policy.

Midjourney

We use Midjourney to generate imagery for marketing and editorial content on the website. We use a Midjourney plan with Stealth Mode enabled so that our prompts and generated images are not made public and are not used to train future Midjourney models. See the Midjourney Terms of Service.

4. The data boundary

We do not send your production data to AI tools. For grading, we send the text of a candidate’s answer to a scenario question — and only that text. We strip identifying metadata before the answer leaves our infrastructure: no name, email address, user or organisation identifier, account history, billing information, or any other field that could link the answer back to a person.

Nothing else from our platform is sent to AI. We do not send AI tools our customer records, organisation profiles, payment data, analytics, session data, or user directories. The same boundary applies when AI assists us in authoring or reviewing questions — the questions themselves are our own content, not customer data.

5. Human review and automation

AI grades scenario responses by default. When an AI-assigned grade falls outside our confidence thresholds — for example, borderline scores, inconsistent rubric signals, or anomalies — a human reviewer looks at the response before the result is recorded. Most candidates receive a grade without that human step.

Once grading completes and a candidate has passed, the certificate issues automatically. We do not have a separate human sign-off on certificate issuance.

We consider this trade-off deliberately: AI grading lets us run the programme at a cost that keeps certification accessible, and the threshold-triggered review is where we think human judgement adds the most value. We will revisit this stance as our grading pipeline matures, and publish changes here rather than quietly shift the bar.

6. Training, retention, and changes

Training. We do not opt in to using our inputs or outputs to train the models of any vendor listed above. The Anthropic API and OpenAI API do not train on API inputs by default, and we have not enabled any option that changes that. Our Midjourney plan (Stealth Mode) is configured so that prompts and generations are not used for training.

Retention. AI vendors may briefly retain inputs and outputs in their own systems for abuse monitoring and operational purposes, in line with their own policies linked above. Where de-identified answer text is sent to the Anthropic API for grading, the grade is recorded in our database; the raw exchange with the API is not retained by us beyond what is required to return a result.

Changes. If we start using a new AI tool, change the data boundary described in section 4, or change the grading and automation posture described in section 5, we will update this page and note the change in the effective date above. Material changes will also be communicated to registered users by email.

Contact. Questions about this policy: hello@ethicsofai.io.