AI Use & Transparency

1. Why this page exists

We use Artificial Intelligence in the core of this product. UK and EU law (UK GDPR transparency duties, the EU AI Act Articles 50 and 86) and ICO guidance say we should explain in plain language what the AI does, what it can't do, what data it sees, and how you can ask a human to step in. This page is that explanation.

This is the AI-transparency notice for the self-service job-seeker product. If you are interacting with our recruiter product as a candidate, you should also receive a candidate-facing notice from the recruiter; the rules described in §6 below ("Human oversight for the recruiter product") are the framework we ask recruiters to operate inside.

2. What model we use and where it runs

• The core AI model is Google Vertex AI Gemini (currently the Gemini 3.1 Flash-Lite family for the self-service product). Vertex AI is a managed service operated by Google Cloud.
• We pin the deployment region to Google's EU multi-region. That means inference (the model running on your prompt) takes place across several Google data centres physically located in EU member states — not in the United States, not in the United Kingdom, and not in a single named EU country.
• We have an automated configuration guard in our codebase that refuses to send traffic to any non-EU region; if a non-EU value appears in our deployment configuration, our test suite fails before release.
• For some advanced or sensitive features (preview only) we may run self-hosted models on our own EU infrastructure (e.g. MLX-served models). When we do this for any feature you can see, we will say so on this page.

3. What the AI does and doesn't do

The AI is used for:

• rewriting and rephrasing CV sections to fit a given job description;
• summarising long CVs or job descriptions;
• generating fit explanations, suggested skills, and tailored cover-letter drafts;
• computing similarity / match scores between a CV and a job.

The AI is not used to:

• make any solely-automated decision about you (about hiring, access to a job, eligibility, or anything else);
• build a behavioural profile of you for advertising;
• infer protected characteristics — we ask the model not to use gender, ethnicity, religion, age, disability, sexual orientation, or close proxies (name, postcode, photo, graduation year), and we strip identifying fields before sharing CVs in the recruiter anonymised-share path.

4. Your data is not used to train Google's models

Under our agreement with Google Cloud — specifically the Google Cloud Data Processing Addendum (Signed — Cloud Data Processing Addendum, Google Cloud, effective 2026-05-20) and the Vertex AI generative-AI service terms — Google does not use the prompts you send to Vertex AI, or the responses it returns, to train its foundation models. Google may briefly cache prompts and responses for abuse detection and to return the response to us; this caching has minimum retention and is governed by Google's published terms.

We also do not use your CV content to train any AI model that we operate. The model you see is a stock Vertex AI model running on your prompt at inference time — there is no "your data, our model" feedback loop.

5. Limitations of the AI

Large Language Models are useful but imperfect. You should expect that AI output can:

• invent facts ("hallucinate") — including skills, dates, employers, certifications, and achievements that are not in your real CV;
• drop or paraphrase important details, sometimes in ways that change their meaning;
• be biased — reflecting biases in the data Gemini was trained on, even though we instruct the model not to use protected characteristics or close proxies;
• be generic, repetitive, or out of date with respect to a job market that has moved since the model was trained;
• produce match scores or fit explanations that are estimates, not authoritative judgements — they are best read as one signal among several.

Always read, fact-check, and edit any AI Output before you rely on it or share it with anyone. You are responsible for what appears in your final document; we are not.

6. Human oversight (recruiter product)

CV ranking and selection in the recruiter product is classified as high-risk AI under Annex III(4)(a) of the EU AI Act. That triggers specific obligations on us as the AI provider and on the recruiter as the deployer:

• No auto-rejection. The fit score is decision-support, not a decision. The recruiter sees the AI ranking next to each candidate but is required to make their own assessment and to give a human reason for advancing or rejecting a candidate.
• Trained reviewers. Recruiters must designate individuals trained to oversee the AI ranking, to understand its limits, and to override it when appropriate.
• Logging. We log reviewer actions (overrides, advances, rejections) for at least six months, so that decisions can be audited and contested.
• Right to contestation and human review. Candidates can ask for a human to re-review the AI output, ask for an explanation of the main factors that shaped the result, and object to AI processing. The recruiter (as controller of the candidate's data) is responsible for routing those requests; we provide a pathway and assist as the processor.
• Bias monitoring. We run an initial adverse-impact test of the scorer before any recruiter beta and re-test on a schedule. Documentation and the latest test results are available to recruiter customers under our DPA.

7. How to ask for an explanation or human review

At any time, for any output of the self-service product or the recruiter product, you can:

• ask us in plain English to explain what the AI did with a specific piece of your data;
• ask us to have a person re-review a result — we will assign a reviewer, not a model;
• ask us to delete a specific AI Output and the prompt that led to it.

Send your request to jobs.privacy@a1ergo.tech with enough detail for us to find the item (e.g. the CV name and the approximate time you generated it). We respond within one calendar month, in line with UK GDPR.

8. Logging and audit

We log AI requests at a minimum level needed to operate the service (timestamp, request size, model used, success/error). We do not store full prompts and responses in our application logs by default. Where we do retain a full record — for example to debug a reported issue, or because a recruiter customer requires audit trails under the EU AI Act — we store it inside the same security perimeter as the original CV and apply the retention schedule in our Privacy Notice.

9. Bias mitigation

Bias in AI systems can produce unfair outcomes for groups protected by the Equality Act 2010 and the EU Charter of Fundamental Rights. We treat that as a serious risk and we mitigate it in layered ways:

• Input controls. We instruct the model never to use protected characteristics or close proxies (name, postcode, photo, graduation year, etc.) as ranking signals. Our anonymised client-shared CV strips these fields before the file leaves our system.
• Process controls. No solely-automated rejection. Human review by trained recruiters in the recruiter product.
• Monitoring controls. Initial and periodic adverse-impact testing of the scorer, with a documented remediation pathway if a disparity is found.

No mitigation is perfect. If you believe the AI has produced an unfair result, please contact us; we will take it seriously and your rights to contestation, explanation, and complaint to a regulator are preserved.

10. Our role under the EU AI Act

When the recruiter product is operated by a recruiter customer:

• We are the provider of the AI system (Article 16). We are responsible for design, risk management, technical documentation, accuracy and robustness, logging, transparency, and cooperating with regulators.
• The recruiter is the deployer (Article 26). They are responsible for using the system in line with our instructions for use, for human oversight in their workflow, for retaining logs, and for informing candidates that an AI system is in use.

We cannot offload our provider duties to Google by virtue of using Vertex AI. We document this allocation in the recruiter Data Processing Agreement.

11. AI-assisted disclosure (self-service)

The self-service job-seeker product is not high-risk under the EU AI Act, but Article 50 still requires us to be transparent that you are interacting with an AI system. This page, the in-app "AI" labels you see on generated output, and the disclaimers on the Terms of Service and your dashboard satisfy that obligation.

12. Changes

We will update this page when we change the AI components of the product (e.g. swap a model, change region, introduce a new feature with new risks). The current version is v1.1, effective 2026-05-20. Material changes will be flagged in-app.

13. Contact

Questions about how we use AI? Want to ask for an explanation or a human review? Email jobs.privacy@a1ergo.tech. You can also complain to the Information Commissioner's Office or your local EU supervisory authority — see the Privacy Notice for details.