As mobile becomes an AI privacy battleground, the risks are suddenly becoming clearer as Google, Samsung, and Apple vie for your attention…
Gemini was first announced as a Copilot-like addition to Workspace Labs, then as a paid Google One addition to Workspace itself, but now it's moved to Gmail on Android and has millions of users. It seems to be reaching the mainstream for users. They also target mobile users, so they operate differently and have very different risks.
AssembleDebug covered Gmail's development beta and wrote, “We've discovered that Google is preparing Gemini AI-powered reply suggestions for Gmail on Android… It's good to see that they're adding to the functionality.'' Assuming this feature is up and running and working on the backend, and if there are no hiccups along the way, we'll be rolling it out to users in the coming weeks. Deployment is scheduled to begin. ”
Desktop beta testers and premium subscribers are attracted to Gemini's offer to “summarize email threads and suggest responses to email threads.” [and] Find information from previous emails [and] Google Drive files. ” The main purpose of the desktop is to filter information and assist in composing emails.
Android apps use Gemini differently. Each email has three suggested replies of varying levels of sophistication that you can select, tap, and send on demand. And the appeal of generative AI continues to grow.
Android screenshots shared via PiunikaWeb, It represents an improvement to the very simple autoresponder that we've all become accustomed to over the last few years. These suggested replies mean Gemini has read your email, even if you haven't read it yet.
The desktop versions of Google Workspace and Microsoft Office AI services have substantial privacy and security considerations. Prompts and user data sent to the cloud, potential storage and human review, use as training data, and data breach risk.
Workspace Labs users are advised that “Google uses Workspace Labs data and metrics to provide, improve, and develop products, services, and machine learning technology across Google. Workspace Labs data may be read by: , rated, annotated, and may even be reviewed.'' Human Critic. ” So users are told, “Do not include personal, confidential, or confidential information in your prompts.”
Google emphasized to me that the Workspace Labs program is an opt-in for people who want to help experiment with and develop new tools. They told me that “only a small portion of our users are participating,” and that users will be given “clear notice” before Google uses your data to improve its services and models. ” was granted and permission was sought.
But mobile will be the real privacy and security battleground for AI.
For Android users, analyzing private email messages to create replies and context is similar to the possibility of Gemini on Google Messages to do the same. This flavor of generative AI is what gets to the heart of the emerging “hybrid AI” debate. Which analytics should be restricted to your local device, and what should be sent to the cloud for more powerful analysis or storage and review.
Privacy details specific to this potential Android release are not yet available, so you can look at what other Gemini applications do for guidance.
Google warns subscription Workspace users that “you may use your Workspace to train or improve, without your permission, the underlying generative AI and large-scale language models that power Gemini, Search, and other systems external to Workspace. We will not use your data.”
But in Gemini's Google Messages, users are told: “Google collects conversations in the Gemini app, related product usage information, location information, and feedback…read and annotated by human reviewers to improve quality and improve our products. , handles conversations for Gemini apps.”
These users are advised not to “enter sensitive information into conversations or enter data that you don't want reviewers to see or that you don't want Google to use to improve its products, services, and machine learning technology.” Please do so (in bold).
While these are clearly different applications from Gemini, such as Gmail, Google Messages, Gemini's own apps, and the broader Workspace service, users trying to understand the privacy implications of using generative AI For this we need to simplify this. Reconciling different privacy policies is difficult. My advice is to differentiate between on-device and off-device and have an additional set of protections if Google has human review and model training elements in place.
On the other hand, the basic advice when sending prompts to cloud-based AI is to follow Google's guidance and avoid any sensitive information. As for the AI in Gmail and similar apps, details are awaited if the data was leaked from users' smartphones, as the information is sensitive in nature.
Google has always been keen to emphasize user consent when it comes to AI, including adding Gemini to its apps, and previously told me: [Gemini] Across Google's products and services, we always uphold this important principle. [Gemini] Always ask users for their explicit consent and notice of data use before accessing their personal data for the first time. ”
We are awaiting details about the scope of Gemini's Gmail rollout on Android and associated privacy and security considerations for its millions of users. And now there's even more spice with the news that Gemini might be coming to her iPhone too. It's unlikely, but almost certain, that Google will analyze your messages within Apple's Mail app or iMessages platform (her Gmail on iOS).
This Gmail update is pre-release, so there is uncertainty regarding timing, functionality, and availability. That said, the beta version of the message I reported was released shortly thereafter.
Google won't comment further at this stage as this is a pre-release, but emphasized its commitment to privacy. That is welcome, but with such apps and services, it is the user's responsibility to understand the risks and use what is available to them with that in mind. When it comes to generative AI, it is important to read the relevant privacy policy, if available.