Google has announced it has made a new privacy tool freely available for all.
Announcing the Magritte tool in a post (opens in new tab) on the Google Developers blog, the company wrote the launch will be the latest addition to Google’s Protected Computing initiative (opens in new tab), which the company claims is to fundamentally change “how, when, and where data is processed to technically ensure its privacy and safety.”
The new tool, which will be available on the open source project repository Github, uses “high-accuracy” machine learning to detect identifying objects, such as licence plates and tattoos, and automatically blur them out.
Google claims that Magritte is best placed to help videographers and video journalists secure the privacy of others in the world around them. It highlighted that it has a low computational spend, and that its high-accuracy makes it a reliable time-saving tool.
Elsewhere, the less catchy “Fully Homomorphic Encryption (FHE) Transplier”, which aims to allow data scientists compute encrypted data without being able to access personal information, and was first released last year (opens in new tab), has received new circuit optimizations to expand its use cases by ensuring a lower computational cost.
The tools are the latest examples of Google focusing on the research and development of privacy-enhancing technologies (PETs) that, in June 2022, became a focus (opens in new tab) for the US Government’s Office of Science and Technology Policy (OSTP).
In November 2022, the concept was also featured as part of a contest (opens in new tab) run by the US and UK governments, which asked participants to develop solutions that allowed for the training of artificial intelligence models without exposing personal data, a principle known as differential privacy (opens in new tab).
In 2019, Google made its differential privacy library – a set of tools designed for ease of use by developers and, in some cases, such as the Privacy on Beam (opens in new tab) privacy framework, absolute non-experts – available on GitHub (opens in new tab).
Be the first to comment