AI Use Policy

• No credential without a human. A Competency Passport is never issued by AI alone.
• Transparency of role. Every AI-graded surface is labelled as such; every reviewer decision is recorded.
• Contestability. Candidates can request a human review of an AI-driven decision that affects them.
• Data minimisation. We send only the fields required for the task to model providers.
• Model provenance. We record which model version produced a given practice score.

The service uses AI in three distinct places, with different guardrails in each:

1. Practice grading (AI-only). When you complete an attempt inside your Development Profile, an AI grader produces a Practice Performance Score against the rubric. This score is clearly labelled as AI Practice Only and is never eligible for a verified Competency Passport. See our Assessment Policy.
2. Reviewer assistance (AI-assisted, human-decided). For a verified assessment attempt, the AI grader produces a first-pass rubric analysis and flags integrity signals for a human reviewer. The reviewer sees the AI output alongside the raw candidate response, adjusts scores, and records a disposition. Only the reviewer-approved score is used for credential issuance.
3. Content generation for beta operations. We use AI to draft simulation prompts and rubric commentary that are then reviewed by our assessment team before publication.

• Deciding whether to issue, revoke, or expire a Competency Passport.
• Identity verification decisions.
• Determining employability, hireability, or fitness for a specific role.
• Automated profiling for advertising, insurance pricing, or credit decisions — we do not conduct these activities.

We use commercial large-language-model APIs (currently the Lovable AI Gateway, which routes to models operated by third-party providers). The data we send is limited to:

• The rubric definition and prompt for the attempt;
• The candidate's response text for that attempt;
• An anonymised attempt identifier — no email, no name, no other profile fields.

We do not authorise providers to train their base models on Competency Labs data, and we prefer providers whose contractual defaults exclude such training. Where a provider changes its data-handling posture, we will update this page and — for material changes — notify affected account holders.

Our AI grader also produces integrity signals (for example, indicators of copy-paste from external sources, unusually generic phrasing, or answers that appear to be machine-generated). These signals are advisory only — they are surfaced to human reviewers, who make the final integrity disposition. A signal is never sufficient, on its own, to revoke a credential.

You have the right to:

• Know when AI was used in a decision that affects you;
• See the reviewer-adjusted score alongside the initial machine score on your Passport (for verified attempts);
• Request a human review of any AI-driven practice score you believe is materially wrong, by emailing appeals@competencylabs.com;
• Withdraw from AI-driven practice at any time without penalty.

We will update this policy when we add a new AI-driven feature, change model provider, or change what data is sent for grading. Material changes are announced on the site and by email to affected account holders.