Responsible AI at NSOffice.AI

Every request that involves an AI-generated response passes through our guardrails layer before it ever reaches an AI model. Guardrails are designed to detect and redact personal data, screen for prompt injection and other manipulation attempts, and check content against safety rules - before generation, not as a filter applied after the fact.

We do not use your organization's chat content, documents, or connected business data to train our own models or any third-party provider's general-purpose models. See our Your Data & Model Performance page for the full detail.

The Service supports multiple AI model providers, so we can match the right model to the right task rather than forcing every request through a single one. Every model - whether from a third-party provider or our self-hosted model - goes through the same governance process before it's used in production:

Because the Service can take actions on your behalf sending a communication, updating a record, triggering a workflow not just generate content, we treat action-taking as a distinct responsibility. Where an action would have an external or irreversible effect, your organization controls the scope of what the Service is authorized to do, and is responsible for reviewing any confirmation step presented before that action executes. We don't design the Service to take consequential actions with no human in the loop.

Our specialized AI agents (“Bees”) - including image generation, video generation, and research tools are built with the same safety screening as the rest of the platform. We prohibit using these tools to create content that impersonates a real individual without consent or that is designed to deceive a third party about its authenticity, as set out in our Usage Policy.

Our guardrail and safety systems are tested on an ongoing basis, including periodic adversarial testing, to confirm they perform as intended across the range of content our platform actually handles not just the cases they were originally designed for.

Our AI governance practices are certified to ISO/IEC 42001, the international standard for AI management systems, covering risk assessment, human oversight, and lifecycle governance for every model we deploy. This sits alongside our ISO/IEC 27001:2022 certification for information security. See our Trust & Compliance page for both.

We aim to be clear about what's AI-generated versus what requires your own review, and about the role human judgment plays at each of the three layers - Think, Know, and Do. The Service is built to make your team faster, not to make decisions on your organization's behalf without your knowledge.