World-class

enterprises trust WRITER

Does WRITER train its models on my data?

WRITER does not train its models on your data, nor does it train on user inputs or any outputs. You control how your data is used.

How does WRITER protect my data?

We maintain and implement various administrative, technical, physical, and organizational security measures to protect the data you share with us. For example, all access to our internal administration tools is logged and periodically reviewed. Additionally, access to user data is restricted and only granted when required for job functions. Any access to user data requires security approval for access. We also support all major SSO providers and multi-factor authentication to help you securely deploy WRITER to your organization, and we use secure keying practices that associate data with customer identifiers to maintain confidentiality. For a WRITER account with multiple teams, inputs and data from one team aren’t used or accessible by another team, even within the same org. For more information about our security practices, including requests to review our SOC 2 (Type 2) report, please reach out to trust@writer.com.

What data does WRITER store?

WRITER stores two main types of customer data:

1. Customer data: company terms, snippets, documents, and style guides.
2. AI inputs and outputs: uploaded media, tuning data, source material, prompts and responses, applications.

WRITER takes a zero data retention approach, storing data for only as long as is needed for you to build with our platform. None of your data is used for training our models. All data, files, and other materials that customers make available to WRITER for using our platform, along with any inputs or generated outputs, are owned by that customer.

What mechanisms are in place to mitigate bias and inappropriate content?

WRITER takes measures to reduce and protect against bias, including: data cleaning and preprocessing, oversight and annotation guidelines, model auditing, fine-tuning from human feedback, and a user feedback loop. We also offer a variety of filters that can be used to remove sensitive data.

We will provide documentation about the mechanisms of our toxic content filters and bias mitigation methodologies upon customer request.

How does WRITER monitor for bias and toxicity?

We’ve developed an independent system that classifies input and output textual data and predicts its level of toxicity. This system also checks for other content attributes, like insults, profanity, identity attacks, threats, and sexually explicit words.

We will provide documentation about the mechanisms of our toxic content filters and bias mitigation methodologies upon customer request.

What measures are in place to protect against prompt injection?

Our platform prevents the rendering of any images or clickable links in output and cannot call out to any URL in the output.

What compliance and security certifications does WRITER have?

WRITER maintains SOC 2 Type II compliance and undergoes annual examinations that evaluate the controls we have designed and implemented based on the trust services criteria relevant to Security, Availability, and Confidentiality. We also maintain certifications for HIPAA Type 1 and PCI compliance and adhere to our obligations under global privacy laws.

How does WRITER comply with the EU AI Act?

We are committed to complying with the EU AI Act as we continue to evaluate our obligations under it and are similarly committed to complying with our obligations under current laws and regulations, in particular those that apply to AI systems and the development of those systems. As laws, regulations, and guidance from regulators continue to evolve, we will continue to follow these developments closely and likewise evolve our approach as needed.