Generative AI Guidelines

Shirabe AI uses generative AI to support research, document understanding, source review, slide generation, spreadsheet/document creation, and narrated video generation.

When the service is used for healthcare or medical information, AI outputs must be treated as informational support only. Users must verify the cited evidence and consult a physician or qualified professional before making clinical or medical decisions.

The Company does not provide medical diagnosis, treatment decisions, legal advice, financial advice, or other professional decisions through AI outputs.

The primary agent execution environment for Shirabe AI is Claude Managed Agents provided by Anthropic.

AI agents run in Anthropic's managed cloud sandbox environment with strong isolation and security controls. Communication outside the sandbox is limited to permitted methods configured for the service.

Files loaded by the AI agent and files generated by the AI agent are not directly accessible from the outside except through limited service-controlled methods.

Inputs, uploaded files, and generated outputs processed by Claude Managed Agents are stored and processed in Anthropic's managed cloud environment.

Data processing may occur globally depending on the infrastructure and external services used by the AI model providers, MCP providers, cloud services, and integration providers.

Data processed by Claude Managed Agents is separated from the Shirabe AI application-side database.

The application side may store account information, chat history, job metadata, usage status, and billing state to the extent necessary to provide the service. Authorization data necessary for MCP or OAuth-based integrations is securely managed in the relevant LLM vendor's cloud environment, not in the Shirabe AI application itself. The service does not store AI model provider API keys in user-accessible records or pass application secrets to AI agents in plain text.

Users must not enter passwords, private keys, access tokens, or other unnecessary secrets into chat prompts or uploaded files.

User inputs, uploaded files, generated outputs, and chat history are not used by the Company to train, retrain, or fine-tune foundation models.

Shirabe AI also uses OpenAI and Google models where necessary for supported features.

AI-generated outputs are produced in response to user instructions and may contain mistakes, omissions, bias, outdated information, or content requiring rights review.

Users are responsible for checking evidence URLs, source documents, citations, licenses, and factual accuracy before relying on or publishing outputs.

The Company may update these guidelines as the service, applicable laws, and AI provider technologies evolve.