PLARO AI Transparency

1. Features Using AI

• Requirement drafts: draft requirements are extracted from documents or notes you upload or paste.
• Meeting and conversation checks: meetings added to a project (their transcripts) and messenger conversations or meeting-app transcripts with your clients are compared with the agreed requirements, to find candidate changes, new requests and open questions, with their evidence.
• Recordings: recordings and meeting recordings are transcribed, speakers are told apart, a title is suggested, and, if you choose, the transcript is translated.
• Meetings themselves run on LiveKit servers that PLARO operates, which are not AI. Background blur in meetings is done inside your browser with a video segmentation model (MediaPipe), downloaded from PLARO’s servers.

2. When AI Runs

• AI runs only for work you start. When a recording ends, its transcription and title (and translation, if chosen) follow; when you add a meeting or a conversation to a project, its comparison with the requirements follows.
• There is no background AI automation that you did not start.
• AI results are stored as drafts or candidates for a person to check. Nothing is added to requirements, documents or tasks, and no e-mail is sent, until you choose and apply it yourself.

3. How AI Results Are Marked

• Results produced by AI are marked on screen as AI results (drafts or candidates).
• When content leaves the service — by copying, Excel export, or the developer AI connection — anything AI wrote is labelled "[AI-written]".
• AI results are not used to make decisions about you (such as sign-up, payment or restrictions).

4. Model Providers

• PLARO uses models from the OpenAI API; when OpenAI cannot respond, a model from the Anthropic API handles the request instead. Recordings are transcribed with OpenAI and AssemblyAI, and when speaker separation is turned on for an in-person recording, AssemblyAI tells the speakers apart. Everyone uses the same models; how much AI you can use depends on your plan.
• AI providers and models may change for quality, cost, reliability, legal or security reasons; if they do, we update this document and the Privacy Policy first. If such a change would materially reduce what your paid plan does, or send your information to a new company or country, we tell you 30 days before (later only for urgent legal or security reasons), and you may cancel and get back the price of the period not yet provided.

5. Limits and Review Duties

• AI outputs may be inaccurate, incomplete or biased, may infer sources incorrectly, or may make up content. Do not rely on facts in AI outputs without checking them yourself.
• Users must verify facts, rights, and legal compliance before publishing, sharing, submitting, executing, or using AI outputs for important decisions.
• If an AI output is inappropriate or harmful, tell us through customer support (hyunjaeyoo@plaro.kr) or the in-app inquiry.

6. Data Processing

• When you use AI features, what you enter, the files, recordings and transcripts you choose, the generated results, model usage and error information may be processed by PLARO and the external model providers. For conversation checks only the words are sent, not names or times.
• AI and transcription providers are used only on terms under which they do not use the information PLARO sends to train their AI models, and information received through the Google Calendar connection is never sent to AI features (Privacy Policy, Sections 6 and 7).
• If you enter a developer’s email address and that developer signs in, the AI coding tool they choose reads the project’s requirements and change history. That tool and its AI company are not providers chosen by PLARO, and what happens next follows that company’s terms and the developer’s account settings (Privacy Policy, Section 4).
• Before entering confidential or sensitive information, review sharing scope and necessity. Avoid unnecessary national identifiers, payment credentials, health data, and authentication secrets.