Build responsible web experiences with generative AI

Alexandra Klepper
Alexandra Klepper

Published: September 16, 2024

Generative AI is everywhere, and the opportunities for developers to create new content and new experiences are endless. With that, it's critical that we build with ethics in mind and create safe experiences for our users. We intend to make AI helpful for all, improving the web ecosystem and user experiences.

Google is committed to responsible AI practices. Whatever decisions you make, be sure to document your practices for the future. Keep a record of the results and best practices so that you and your colleagues can continue to maintain and improve existing and future applications.

Sites, apps, and extensions that generate content using AI should consider the following:

  • Rigorously test AI tools and models.
  • Use pre-existing safety tools and settings.
  • Give users a way to report or flag offensive content, then address those reports.
  • Advertise your product responsibly.

Rigorously test AI tools and models

You or your company are accountable for the experience in your apps, so it's critical that you understand the underlying AI models and technology. It's important to test a variety of use cases and safeguard against prompts that could create harmful or offensive content.

Use pre-existing safety mechanisms

Some APIs, such as Vertex AI, have tools and configuration options for safety filters. Safety filters act as a barrier to prevent harmful output, but they don't directly change the model's behavior. We recommend you review and set these filters to a level you deem appropriate.

Enable offensive content reports

Even with testing and safety filters, it's possible that content that's considered to be offensive content could make its way through to users. Monitor and prioritize user feedback around generative AI features, to ensure you're shaping the best possible experience. Establish a process to address reports of offensive content.

To mitigate potential offensive content, you could perform one or more of the following actions:

  • Block similar inputs that led to the offensive content
  • Redact the model output
  • Fine tune the model
  • Switch the model entirely

If you build browser extensions, always follow the relevant policies. Chrome Extension developers who would like to publish their extension on the Chrome Web Store should follow the Chrome Web Store policies. If you publish content generated with AI, aim to create helpful, reliable, and people-first content.

When advertising, it's important to do so in a way that's safe and respectful of users. Review your marketing materials to ensure that your ads accurately represent your site or app's actual capabilities.

Looking to the future

We'll continue to provide best practices and share resources, such as the People + AI Guidebook, to support developers in building innovative and responsible apps.

Share your feedback with us at @ChromiumDev on X or Chrome for developers on LinkedIn, and explore all of our AI content.