# Responsible AI ยท Communicate > How Communicate builds responsible AI: grounded answers, human oversight, privacy by default, and no training of foundation models on your data. - **URL:** https://communicate.so/responsible-ai --- - [Home](/) - - Legal # Responsible AI Legal reference Last updated: July 22, 2026 Communicate is an AI customer-support platform operated by TinyCheque, Inc. This page explains how we build and run our AI responsibly, where its limits are, and what we ask of the businesses that use it. We would rather be honest about what the technology can and cannot do than make claims we cannot stand behind. On this page - [Our approach to responsible AI](#our-approach-to-responsible-ai) - [Principles](#principles) - [How the AI agent works](#how-the-ai-agent-works) - [Human oversight and takeover](#human-oversight-and-takeover) - [Accuracy, limitations, and hallucinations](#accuracy-limitations-and-hallucinations) - [Data and model use](#data-and-model-use) - [Privacy and security](#privacy-and-security) - [Fairness and non-discrimination](#fairness-and-non-discrimination) - [Transparency and disclosure](#transparency-and-disclosure) - [Prohibited and high-risk uses](#prohibited-and-high-risk-uses) - [Governance and accountability](#governance-and-accountability) - [Reporting concerns and contact](#reporting-concerns-and-contact) ## Our approach to responsible AI Communicate helps businesses answer their customers with an AI support agent that is trained only on the customer's own knowledge base and connected data sources. A shared inbox lets human agents step in and take over from the AI at any point. That combination, an AI grounded in your own content plus a human always able to intervene, is the foundation of how we think about responsible AI. We do not treat AI as a black box that replaces people. We treat it as a tool that drafts and delivers answers quickly, under human supervision, using content the business has chosen and controls. Our goal is to be useful without overpromising. AI language models are powerful, but they can be wrong, incomplete, or confidently mistaken, and a responsible product is designed around that reality rather than pretending it away. The principles below describe the commitments we hold ourselves to. They are not marketing language. Each one maps to how the product actually works, and where a principle depends on the business using Communicate rather than on us, we say so plainly. ## Principles Six principles guide how we design, ship, and operate the AI in Communicate. - Grounded answers. The AI agent answers from the customer's own configured knowledge base and data sources, not from open-ended general knowledge. Answers are meant to reflect what the business has actually written and approved. - Human oversight. A human agent can take over any conversation at any time through the shared inbox. Escalation from AI to human is a core feature, not an afterthought. - Transparency. We are candid about what the AI is, how it works, and where it can fail. We encourage businesses to be equally candid with their own end users. - Privacy by default. Customer data and end-user conversations are used to run the service, not to train third-party foundation models. Details live in our Privacy Policy at /privacy. - Fairness. We aim for the AI to treat end users consistently and to avoid unfair or discriminatory handling, while being honest that language models can reflect biases present in their underlying training. - Accountability. TinyCheque, Inc. is responsible for the platform. Businesses are responsible for the knowledge they configure and for how they deploy the AI to their end users. ## How the AI agent works When a business sets up Communicate, it connects its own knowledge sources, such as help articles, documentation, product information, and other approved content. The AI support agent uses that content as the basis for its answers. It is designed to respond from what the business has configured rather than to improvise from general world knowledge. Inference, meaning the actual generation of each reply, runs on gpt-4o-mini accessed through OpenRouter. We use the model for inference only. We send the relevant context and the end user's question to the model, and it returns a drafted answer that is grounded in the retrieved knowledge. The quality and accuracy of answers therefore depend heavily on the quality, accuracy, and completeness of the knowledge the business provides. Because answers are grounded in a specific, business-controlled knowledge base, the AI is narrower and more predictable than a general-purpose chatbot. That is deliberate. A grounded agent that says less but stays closer to approved content is more trustworthy in a support setting than one that answers everything from general knowledge. ## Human oversight and takeover Every Communicate workspace includes a shared inbox where human agents and the AI work side by side. A human can read any conversation, jump in, and take over from the AI at any moment. Once a human takes over, they, not the AI, are handling the conversation. This matters because no AI system is right every time. Human takeover gives businesses a direct, always-available way to correct, clarify, or handle anything the AI should not resolve on its own. Sensitive, unusual, or high-stakes conversations can and should be routed to a person. - Humans can monitor AI conversations in real time through the shared inbox. - Humans can take over a conversation at any point, and escalation is a core, built-in capability. - Businesses decide how much they rely on the AI versus their own team, and we recommend keeping a human in the loop for anything consequential. ## Accuracy, limitations, and hallucinations We want to be direct about this: the AI can be wrong. Even when grounded in a customer's knowledge base, language models can produce answers that are inaccurate, incomplete, outdated, or plainly incorrect, sometimes while sounding confident. This behavior, often called hallucination, is a known limitation of current AI technology, and Communicate is not immune to it. The single biggest factor in answer quality is the knowledge the business configures. The AI can only be as accurate as the content it is grounded in. If that content is wrong, ambiguous, or missing, the AI's answers will reflect those gaps. Each business is responsible for the accuracy, completeness, and currency of its own knowledge base and data sources, and for keeping them up to date. - Review and maintain your knowledge base so the AI is grounded in accurate, current content. - Use human oversight and takeover for anything where a wrong answer would cause real harm. - Do not rely on the AI as a sole source of truth for critical decisions without human review. - Disclose to your end users that they may be interacting with an AI, so they can weigh answers accordingly. We do not publish accuracy percentages, benchmark scores, or red-team results here, because we will not present numbers we cannot honestly stand behind for every business and every knowledge base. Real-world accuracy depends on your content and your configuration. ## Data and model use Communicate does not use customer data or end-user conversations to train third-party foundation models. The model we rely on, gpt-4o-mini via OpenRouter, is used for inference only. We send context to generate a reply; we do not hand your data over to be absorbed into the training of a general-purpose model. This is a firm commitment, not a preference we might quietly change. When we say inference only, we mean the model is used to produce answers in the moment, not to learn from your data for someone else's benefit. How we collect, use, retain, and protect data across the platform is described in full in our Privacy Policy at /privacy. That policy is the authoritative reference for data handling, and this page should be read alongside it. ## Privacy and security Responsible AI is not only about the model. It is also about how the surrounding data is handled. We build Communicate to keep account data and end-user conversation data separated by purpose and used only to operate the service the business pays for. Our detailed data practices, including what we collect and the choices you have, are in our Privacy Policy at /privacy. Our security practices and the specific measures we take to protect data are described on our security page at /security. We keep security specifics there so they stay accurate and in one place. We want to be clear about what we do not claim. We do not claim any formal certification we do not hold. This page makes no claim to SOC 2, ISO 27001, or HIPAA compliance. If our certification status changes, we will say so on our security page rather than imply it here. ## Fairness and non-discrimination We want the AI to treat end users consistently and fairly, and to avoid unfair or discriminatory treatment based on personal characteristics. Because answers are grounded in a business's own knowledge base, much of what the AI says is shaped by the content that business provides. At the same time, we are honest that AI language models can reflect biases present in their underlying training data, and no provider can guarantee that every output is free of bias. We ask businesses to review how the AI responds across different users and scenarios, to keep their knowledge base free of discriminatory content, and to use human oversight where fair and even-handed treatment is essential. ## Transparency and disclosure We believe end users deserve to know when they are talking to an AI. We strongly recommend that businesses using Communicate disclose to their own end users that they may be interacting with an AI support agent, and that a human can take over when needed. Clear disclosure builds trust and sets accurate expectations. It helps end users judge answers appropriately and know that they can reach a person. Depending on where a business operates, disclosing the use of AI may also be a legal or regulatory requirement, and each business is responsible for meeting the obligations that apply to it. - Tell your end users, in plain language, that they may be chatting with an AI. - Make it clear that a human agent can take over the conversation. - Keep your disclosures consistent with the laws and regulations that apply to your business. ## Prohibited and high-risk uses Communicate is a customer-support tool. It is not a substitute for professional advice, and it should not be positioned as one. In particular, the AI must not be presented to end users as a source of professional medical, legal, or financial advice. The AI should also not be used to make fully automated decisions that produce legal or similarly significant effects for a person without meaningful human review. Decisions of that kind belong with a human who can be accountable for them. - Do not use the AI as a replacement for licensed professional medical, legal, or financial advice. - Do not rely on the AI for fully automated decisions that have legal or similarly significant effects on people without human review. - Do not deploy the AI in ways that would deceive, manipulate, or harm end users, or that violate applicable law. - Do not configure the AI with content intended to discriminate against or exploit end users. Where the stakes are high, keep a human in the loop and use the takeover and escalation features that Communicate provides. ## Governance and accountability TinyCheque, Inc. operates Communicate and is accountable for the platform, including the choices we make about how the AI is built and run. We take responsibility for the parts of the system we control: the product, the grounding approach, the inference setup, human oversight tooling, and our data practices. Responsibility is shared. Businesses that use Communicate are accountable for the knowledge base they configure, for how they deploy the AI to their end users, for disclosing AI use where required, and for keeping a human in the loop for consequential interactions. Responsible outcomes depend on both sides holding up their part. As the technology, our product, and the surrounding regulations evolve, we will revisit this page and update it. The date at the top reflects when it was last changed. ## Reporting concerns and contact If you have a concern about how the AI in Communicate is behaving, a question about our responsible AI practices, or a report of misuse, we want to hear about it. Prompt, honest reporting helps us improve the product and address problems quickly. You can reach us at [communicate@support.communicate.so](mailto:communicate@support.communicate.so). Communicate is operated by TinyCheque, Inc., Newark, Delaware, USA. For data-specific questions, our Privacy Policy at /privacy is the authoritative reference, and our security practices are described at /security.