Google has presented Private AI Compute, its cloud infrastructure specially designed so that our conversations remain totally private and cannot be accessed by anyone else. Not even Google.
Why is it important. The deployment that Google has announced will allow users of models like Gemini to use them without fear that their sensitive data – finances, health, private conversations – could end up being rescued and accessed by third parties.
Idea copied from Apple. This type of infrastructure is an adaptation of the platform that Apple presented more than a year agoPrivate Cloud Compute, and that precisely focused on protecting those conversations by using the company’s future AI models. There are some differences, and for example Apple makes use of a concept of “verifiable transparency” that allows external researchers to audit security and privacy at any time cryptographically. At Google they use third-party verification, which is somewhat more limited as it is not open to the public to verify the running software.
Tranquility as a sales argument. AI models are becoming more useful and also more personal and proactive, and that means that we also end up using them with data that may be more personal and sensitive to help us with a very specific question.
Beyond ZDR. The problem is that when using the models everything we ask and they answer you can see —and even deduce—. There are ZDR (Zero Data Retention) modes in enterprise accounts from some AI providers, but having a cloud that “privatizes” those conversations is especially promising when it comes to being able to talk about everything with AI without restrictions. no fear of that data coming out of there.
How this “privatization cloud” works. Those responsible for Google they explain that Private AI Compute is a “secure, fortified space for processing your data that keeps your data isolated and private to you.” The system uses several layers involving its TPUs and its Titanium Intelligence Enclaves (TIE) security chips. Our devices connect to that secure cloud environment through encryption and a cryptographic security mechanism called “remote attestation” that verifies the identity and integrity of that hardware environment to which we connect. Google also offers a detailed technical report on the operation of this infrastructure.
Similar to running local models. The result is theoretically that for the user everything runs “locally” in terms of privacy. Features such as translation or audio summaries that Google offers in its services run directly on our devices: there is no data that travels to the cloud.
The best of both worlds. The problem is that local AI models have limited performance, and Private AI Compute will allow you to have the best of both worlds: the power of the best AI models—which run in gigantic data centers—and the privacy guarantee of Google’s Private AI Compute.
A surprising twist. This type of infrastructure means that these conversations are completely protected and that not even Google can access them. It’s a surprising turn of events, especially since for the last 25 years Google has made a living by collecting our data to apply it to its advertising model. This type of option goes in just the opposite direction, and it only remains to be seen how it will market such capability.
Strategic approach. Curiously, this announcement comes days after we learned that the new version of Siri with AI It will be powered—at least, initially—by Gemini, Google’s AI model. Both companies have had a multimillion-dollar agreement for years to make Google the default search engine in Safari on iPhones and Macs, and now that alliance is apparently reinforced with the use of Google’s AI model to power the future version of Siri.
In Xataka | The key to making the iPhone competitive in AI was right next door: imitating what Android had already done
GIPHY App Key not set. Please check settings