Skip to main content

    AI Transparency, Human Review & Limitations Notice

    How Agent Smith uses AI across the website and platform, what that means for you, and why human review remains important.

    Legal Notices — index of all published notices.

    Effective date: 2026-08-16 · Last updated: 2026-08-16

    Overview

    Agent Smith includes AI-powered features across parts of the website and platform. These features may support conversations, drafting, suggestions, summarization, content generation, and, depending on workspace settings, certain background or semi-automated workflows.

    This notice explains, in plain language, how those AI features should be understood and used.

    AI features in Agent Smith

    AI may be used in different parts of the Service, including:

    • public website experiences, such as the website assistant or voice transcription
    • authenticated workspace experiences, such as chat, drafting, suggestions, summaries, and workflow support
    • optional background or semi-automated actions where enabled by workspace configuration

    Not every workspace or feature will use AI in the same way, and some AI capabilities may be limited, gated, or configurable.

    You may be interacting with AI

    Some parts of Agent Smith involve interaction with AI systems rather than a human operator.

    Chat, assistant, and agent surfaces are identified as AI at the point of interaction through agent/model labels and assistant presentation. This page supplements those in-product disclosures; it does not replace them.

    Human review matters

    AI outputs can be incomplete, misleading, outdated, or simply wrong.

    You should review and verify outputs before relying on them, especially for decisions involving:

    • legal matters
    • tax or accounting matters
    • finance
    • hiring or HR
    • compliance
    • contracts
    • other high-impact business decisions

    Agent Smith is designed to assist, accelerate, and structure work. It is not a substitute for professional judgment, qualified advice, or independent review.

    No guarantee of accuracy or fitness

    We do not guarantee that AI-generated outputs will be accurate, complete, reliable, or suitable for your intended use.

    This is especially important during beta, where features may evolve quickly and system behavior may change over time.

    Synthetic and AI-generated content

    Some content generated through Agent Smith may be synthetic or AI-assisted.

    Within chat, AI responses carry a machine-readable assistant role and are presented as assistant/model output. Asset workflows preserve an AI-generated flag where that feature supports it. Copying, downloading, or publishing content can remove this context, so users must preserve or add an appropriate AI disclosure when required. Universal watermarking is not implemented for every output type, and users must not infer human authorship from the absence of a watermark.

    AI providers and data sharing

    Agent Smith may use one or more third-party AI providers to deliver AI functionality.

    Active inference providers may include:

    • OpenRouter (model routing)
    • OpenAI
    • Anthropic
    • Google (Gemini)
    • xAI (Grok)
    • NVIDIA (Nemotron)
    • DeepSeek

    When you use AI-powered features, the content you submit, including prompts, files, messages, or audio, may be processed by one or more of these providers in order to generate a response or complete the requested function.

    We do not train our own models on your prompts or files.

    Non-free text requests routed through OpenRouter require Zero Data Retention and deny provider data collection. The NVIDIA Nemotron Free route is an explicit exception: its upstream notice permits logging and use for NVIDIA product improvement. Direct-provider fallbacks follow the selected provider's API terms and do not inherit OpenRouter's per-request ZDR setting. For more information, please refer to the Privacy Policy and Subprocessor List.

    Automated, semi-automated, and approval-based behavior

    Depending on the feature and the configuration of a workspace, Agent Smith may support:

    • AI suggestions that require user review
    • drafts or proposed changes awaiting approval
    • semi-automated workflows
    • background actions or system-driven behavior where configured

    In many cases, the product is designed to request confirmation before meaningful actions are taken. In some cases, however, certain actions may run automatically depending on workspace settings, permissions, enabled integrations, or product configuration.

    You are responsible for choosing settings that fit your risk tolerance and for reviewing actions where appropriate.

    Use AI features carefully

    You should avoid using AI features in a way that creates unnecessary risk, including by submitting content you do not want processed through supported AI providers.

    If you use Agent Smith for sensitive or consequential work, you should apply appropriate internal controls, review practices, and human oversight.

    Changes over time

    Agent Smith is evolving. Our AI features, provider mix, review flows, technical controls, and disclosure practices may change over time, especially during beta.

    When those changes materially affect this notice, we will update this page.