In a new significant milestone for the integration of artificial intelligence (AI) into personal devices, Microsoft has announced substantial improvements in the use of AI in the PC Copilot+. These improvements include the incorporation of a Neural Processing Unit (NPU) of over 40 TOPS, which promises to reduce latency, improve battery life in AI-intensive tasks, enable the use of offline AI experiences, and enhance privacy by keeping information locally.
Microsoft acknowledges that moving AI models and data processing to PCs creates unique security challenges. To address these challenges, the company has introduced “Recall,” a new exclusive experience for the PC Copilot+, designed to help users instantly and securely find what they have viewed on their PC.
The security and privacy design of Recall is based on four fundamental principles: user control, encryption of sensitive data, isolation of services operating on snapshots, and user intentionality in using Recall. During the setup of the PC Copilot+, users can choose to activate Recall and instant snapshots are only saved if this option is chosen. Additionally, sensitive data in Recall is always encrypted and protected by keys, while associated services operate within a secure VBS enclave, ensuring that only the information required by the user is accessible during active use of the function.
The security model of Recall uses VBS enclaves to protect snapshots and associated data. These enclaves segment the computer memory in a protected area where information can be processed securely, using cryptographic attestation protocols to ensure that the environment is secure before performing sensitive operations. Windows Hello biometric credentials also play a crucial role in user protection and authentication.
Recall always prioritizes user privacy. Snapshots and associated data are stored locally on the device and are not shared with Microsoft or third parties without the explicit consent of the user. Users have full control over this functionality, being able to delete, pause, or disable snapshots at any time.
The architecture of Recall includes key components such as secure configurations, semantic indexes, snapshot storage, and user services, all designed to operate within a VBS enclave, reducing the risk of data leakage and protecting encryption keys.
To ensure a meticulous and secure approach, Microsoft has conducted comprehensive security assessments, including design reviews and penetration testing conducted by the Microsoft Offensive Research and Security team (MORSE) and an external provider. Additionally, a responsible AI impact assessment was completed, covering risks, damages, and mitigation analysis based on fair AI principles, security and privacy, inclusivity, transparency, and accountability.
In conclusion, the secure design of Recall provides robust controls against known threats, reaffirming Microsoft’s commitment to making the power of AI accessible to all, while maintaining security and privacy even in the face of the most sophisticated attacks. Through collaboration with OEMs, app developers, and other ecosystem actors, Microsoft continues to move towards a Windows experience that is more secure by design and by default.
Referrer: MiMub in Spanish