Responsible AI.

validant.ai sells trust. So we hold ourselves to the same standards we audit others on. This page is our public account of how we develop, deploy and operate AI: for the features inside our own platform and for the AI we recommend to customers. It is aligned to the EU AI Act (Regulation 2024/1689), the OECD AI Principles, the NIST AI Risk Management Framework and the ISO/IEC 42001 management system for AI.

Three commitments anchor everything below.

• People stay in the loop. No AI system we build or operate makes a decision with legal or similarly significant effect on a person without a meaningful human review.
• Data dignity. We do not use customer content to train general-purpose foundation models. We do not silently re-purpose data. Where data is used to improve our own narrowly scoped models, it is contractually licensed for that purpose.
• Explainable to the person affected. If an AI output influences a decision about you, you have the right to know that it did, what it considered, and to challenge it.

We classify every AI component we build or deploy by risk class under the EU AI Act and publish the classification before customers commit.

Design

Every AI feature starts with a written purpose, a documented set of intended users, an explicit list of foreseeable misuses, and a fairness scoping note. A Data Protection Impact Assessment (DPIA) is started before code.

Data

Training, validation and test data are described in a data sheet: provenance, licence, known biases, exclusions, retention. We do not scrape personal data from third-party sources without a legal basis.

Build

Models are versioned and accompanied by a model card with capabilities, limitations, evaluation metrics and intended use. We track model lineage end-to-end.

Evaluation

We evaluate against fairness, robustness and safety metrics appropriate to the use case. For decision-relevant systems, we publish disparate-impact metrics across protected characteristics relevant to the deployment context.

Deploy

Human-in-the-loop, human-on-the-loop or human-out-of-the-loop is decided per feature and documented. Default for any decision affecting a person is human-in-the-loop.

Monitor

Post-deployment we monitor drift, fairness metrics and incident rates. Any serious incident as defined by AI Act Art. 73 is reported to the competent authority within the statutory deadline.

Decommission

When a model is retired, we document why, archive the artefacts, and keep the audit trail for the legally required period.

A precise account of where customer data does and does not go.

What we never do

• Send your Customer Content to third-party model providers in a way that allows them to train on it
• Use your prompts, audit inputs or platform telemetry to train general-purpose foundation models
• Buy data sets of personal information from data brokers

What we may do, only with your explicit licence

• Use a clearly identified subset of de-identified data to improve a narrowly scoped model that benefits you directly (e.g. fairness-audit calibration). This is opt-in and revocable.

• Anywhere on the platform where you interact with an AI system you see a clear indicator before the interaction begins.
• Synthetic outputs (generated text, images, audio, video) carry machine-readable metadata identifying them as synthetic, following the C2PA Content Credentials standard where applicable.
• We publish a public registry of high-risk AI systems we operate (planned, alongside the EU AI database).

If an AI-supported process at validant.ai produces an output about you, you have:

• The right to be told that AI was involved and, in plain language, how
• The right to a meaningful explanation of the factors that contributed to the output
• The right to a human review
• The right to contest the outcome
• The right not to be subject to a solely automated decision with legal or similarly significant effect (GDPR Art. 22)

Exercise any of these rights at privacy [at] validant [dot] ai.

• An AI Review Board (engineering, legal, ethics) reviews any new AI-relevant feature before it ships, and any feature change that crosses a risk-class boundary.
• All staff working on AI complete annual training on the EU AI Act, GDPR Art. 22 and our internal AI policy.
• Suppliers offering AI services are vetted with a written AI-specific addendum to our DPA, including a no-training guarantee and a model-versioning notification clause.
• Incidents and near-misses are reported into a single incident register reviewed quarterly by the executive level.

How does validant.ai align with the EU AI Act?

Every AI component is classified by risk class under the EU AI Act. Prohibited systems are refused; high-risk systems get a full risk-management system, data governance, technical documentation, logging, human oversight and ISO/IEC 42001-aligned quality management.

Does validant.ai use customer data to train models?

No. validant.ai does not use customer content to train general-purpose foundation models and does not silently re-purpose data. Any data used to improve its own narrowly scoped models is contractually licensed for that purpose.

What is responsible AI?

Responsible AI is the practice of building and operating AI systems that keep humans in the loop, protect data dignity, and remain explainable to the people they affect, governed against frameworks like the EU AI Act, NIST AI RMF and ISO/IEC 42001.