AI is rapidly evolving, and its development is advancing at an unprecedented pace. As we continue to learn from the experiences of those using our tools, we recognize the immense potential of this AI revolution while acknowledging the importance of a responsible and thoughtful approach. At Microsoft, we aim to harness the full power of AI Copilots to create valuable and transformative experiences for users while ensuring the highest levels of safety and accountability.
To strike the right balance between innovation and responsibility, we are introducing Copilot Labs. This initiative allows us to test our most cutting-edge features with a select group of users before a broader release. By gathering feedback from this limited trial, we can refine and enhance the tools, making them both more powerful and safer for all users. Available to Copilot Pro subscribers, Copilot Labs offers a sneak peek into groundbreaking projects that are still in progress.
The first feature launching in Copilot Labs is Think Deeper, which enables Copilot to reason through complex problems. Leveraging the latest reasoning models, Think Deeper can assist with anything from solving intricate math equations to analyzing the costs of home projects. It takes extra time to provide detailed, step-by-step responses to challenging questions. Starting today, Think Deeper will be available to a limited number of Copilot Pro users in Australia, Canada, New Zealand, the United Kingdom, and the United States.
Next on the horizon is Copilot Vision.
One limitation of traditional Copilot tools is their inability to perceive or understand what you’re looking at. While language is a powerful form of communication, many tasks involve context that goes beyond words. Copilot Vision addresses this by offering users the option to let Copilot see what they see. Currently being trialed in Copilot Labs, this feature takes Copilot’s capabilities to a new level.
With Copilot Vision, integrated into Microsoft Edge, the tool can understand the webpage you’re viewing and offer contextual help based on its content. Whether you’re navigating a task, asking questions, or seeking suggestions for your next steps, Copilot Vision responds in natural language. This enhances the ease and practicality of using Copilot across various tasks.
User control and privacy have been paramount in the development of this feature. Copilot Vision is an opt-in service, activated only when you choose. Users decide when, how, and if they want to use it. During this preview phase, none of the content—including audio, images, text, or your interactions with Copilot—will be stored or used for training. Once you close the feature, all associated data is permanently discarded.
For added security and responsibility, Copilot Vision currently avoids paywalled and sensitive content. Initially, its usage is limited to a pre-approved list of popular websites that serve a wide audience. When interacting with these sites, Vision respects their AI-related controls. Over time, we plan to expand this list, always prioritizing safety and user trust. Importantly, Copilot Vision doesn’t engage with the web directly—it’s designed to answer questions rather than perform actions.
To learn more, explore our detailed Q&A, where we address the safety measures built into Copilot Vision and Copilot Labs. Throughout the development of these features, we’ve prioritized the needs and interests of users and creators, ensuring a balance between utility and responsibility. As this experimental feature rolls out to a select group of Pro users in the United States, we eagerly await your feedback to continue refining and improving it.
We hope you find these new capabilities useful, and we’re excited to hear your thoughts as we shape the future of AI together!
