What AI Onyatrips actually uses today, and the principles guiding what we build next.
Last updated: 8 August 2026
Effective Date: 8 August 2026
Last Updated: 8 August 2026
This Policy explains how Onyatrips approaches artificial intelligence and automated processing: what we use today, and the principles that will guide anything we build in the future. We aim to describe our current systems accurately rather than in aspirational terms, and to set out the governance framework we commit to applying as AI capability develops.
This Policy describes how Onyatrips approaches artificial intelligence and automated processing — both what exists today and the principles that will govern anything we build in the future. It applies to any feature on the Platform that uses AI, machine learning, or automated decision-making, including recommendations, matching, moderation, and fraud detection.
To be precise:
The Terms and Conditions previously described destination and companion discovery using language such as "AI matching" and a "hybrid artificial-intelligence and human-editorial process," even though this functionality operates through deterministic, rule-based filtering rather than a machine-learning system. That language has since been corrected in the Terms and Conditions to accurately describe the current, deterministic implementation, with this Policy noted as the controlling, current-vs-future source on any AI-related question. The term "compatibility scores" is retained in the Terms as a label for automated discovery signals — it does not, by itself, imply machine learning or predictive AI; Section 2 reflects the current state of the product regardless of terminology used elsewhere. As we introduce genuinely AI-driven features, we will update our policies to accurately describe the technology in use at that time.
If and when Onyatrips introduces machine-learning-based recommendations, matching, moderation, or any other AI capability with a meaningful effect on Users, we commit to designing it around the following principles, described in the sections below: human oversight, transparency, explainability (to a reasonable degree), fairness testing, data minimization, security, and a functioning complaint/correction pathway. These are forward-looking commitments describing how we intend to build AI responsibly — they do not describe a system that exists today, except where Section 2 says otherwise.
Any future AI system with a significant effect on a User (for example, an automated moderation decision that suspends an account, or a scoring system materially affecting discovery visibility) will be designed with a path to human review, and will not be the sole, unreviewable basis for a decision with a significant effect on a User's access to the Platform, consistent with the automated-decision-making commitments in the Privacy Policy.
Where AI-generated or AI-assisted content is shown to Users (for example, an AI-assisted destination summary, if and when such a system exists), we intend to label it as such rather than presenting it as unassisted human editorial content or as another User's independent opinion, consistent with the Disclaimer.
We do not currently operate an AI system requiring an explainability interface (since none is in production with user-facing effect). Going forward, where feasible, we intend that AI-driven suggestions or classifications communicate an appropriate degree of uncertainty (for example, presenting a "suggested match" as a signal rather than a certainty) rather than false precision, consistent with the Terms and Conditions' framing of compatibility scores as "probabilistic signals only."
Any AI or machine-learning system, including a future recommendation, matching, or moderation system, can be wrong: it can produce outdated information, miss context a human would catch, reflect patterns in its training data that do not generalize well, or be manipulated (see Section 12). We do not, and will not, represent that an AI system's output is guaranteed to be accurate, complete, or fair. The one AI/ML-adjacent system that exists today — the experimental profanity/spam classifier described in Section 2 — inherits this limitation like any classical ML text classifier: it can produce false positives (flagging acceptable content) and false negatives (missing genuinely harmful content), which is a documented reason it is not yet relied upon in the live chat path.
Because no consequential AI/ML system is currently in production, we have not conducted bias testing across protected characteristics (gender, religion, caste, ethnicity, disability, sexual orientation, or similar) for a recommendation or matching system, since none exists to test. Before any future AI system that could differentiate outcomes across Users (for example, a matching or ranking algorithm) is deployed, we commit to assessing it for disparate impact across such characteristics as part of our development process, and to monitoring for bias on an ongoing basis once deployed, consistent with the non-discrimination principles reflected in the Community Guidelines.
We do not currently have a documented AI model-evaluation, safety-testing, or red-team-testing process, since no consequential AI/ML model is currently in production. This will be established before any such system is deployed.
Any current or future AI system is subject to the data-minimization, purpose-limitation, and lawful-basis principles set out in the Privacy Policy. We do not use Personal Data to train, fine-tune, or improve any third-party or generative AI model today, because we do not currently operate or integrate such a model. If we do so in the future, we will disclose it in the Privacy Policy and this Policy, and will seek any consent required by applicable law before using User-provided Personal Data for model training.
The experimental content-moderation classifier described in Section 2 is, like any text-classification system, potentially subject to adversarial evasion (users deliberately crafting text to bypass keyword/classifier detection) — this is a known, general limitation of this class of system, not a claim that our specific deployment has been compromised. Any future generative-AI feature (for example, an AI chat assistant) would need to be assessed for risks such as prompt injection and manipulation before deployment; no such assessment has been performed because no such feature currently exists.
We do not currently integrate any third-party generative-AI provider (such as an LLM API) integrated into the Platform. If one is introduced in the future, we will apply vendor-governance review (data handling, security, and reliability) before integration, and will disclose the provider in the Privacy Policy and this Policy.
We do not currently generate AI images, AI video, or AI text content on the Platform. Rules governing User-submitted AI-generated or synthetic content (for example, an AI-generated photo a User uploads to their profile) are set out in the "AI-Generated and Synthetic Content" clause of the Terms and Conditions' User-Generated Content module, which requires such content to comply with the same standards as any other User content and prohibits deceptive synthetic content (deepfakes, impersonation, or misrepresentation of real people, places, or events).
As described in Section 2 and the Community Guidelines, our experimental moderation classifier is not currently active on the live chat path. If and when it (or a successor system) is activated, we commit to: providing a path for a User to appeal a moderation action they believe was a false positive, consistent with Section 9 of the Community Guidelines; and monitoring false-positive/false-negative rates as part of ongoing evaluation, consistent with Section 10.
Consistent with the Privacy Policy, we do not currently perform solely automated decision-making producing legal or similarly significant effects on Users. If this changes, we will provide the rights required by applicable law, including the ability to request human review of, and to contest, such a decision.
If you believe an AI or automated system on the Platform has produced an inaccurate or harmful result about you (for example, an incorrectly applied moderation flag once the classifier is active), contact us using the details in Section 21. We do not currently have a dedicated AI-incident intake process distinct from our general feedback/support channel; establishing one is a priority for us once a consequential AI system is in production.
We use the EU AI Act's risk-based categorization approach as a design benchmark, not as a law we claim to be certified against, since we do not currently operate the kind of AI systems that Act primarily targets. We also monitor general principles emerging under India's evolving digital-governance framework, GDPR (regarding automated decision-making and profiling), and CCPA/CPRA (regarding automated decision-making technology disclosures), as described further in the Legal Compliance Framework.
We do not currently have a named internal AI governance owner or an AI risk-assessment process, since no consequential AI system is in production. We intend to establish both before any AI system with material effects on Users is deployed.
We will update this Policy as our AI capabilities evolve. Material changes will be reflected in an updated "Last Updated" date above.
We use analytics cookies to understand how you use Onyatrips, so we can improve it. These are optional — the site works fully without them. Learn more.