Google confirming attacks on Gmail users is headline news (1,2). But while these impact a small number of users, there’s a new warning that does affect you and your account. Google is offering upgrades with serious implications if you make the wrong choice.

All the recent headline Gmail attacks abuse Google’s legitimate infrastructure to either bypass password and two-factor authentication (2fA) security or mimic Google’s own employees to trick users into opening up their own accounts to attackers.

Google recommends all users upgrade their accounts to add passkeys and then to use these instead of passwords and 2FA. But that’s not the only upgrade for Gmail users. The others are not as straightforward and come with a new warning.

We’re talking AI, of course, and a new warning from Incogni that “platforms developed by the biggest tech companies turn out to be the most privacy invasive, with Meta AI (Meta) being the worst, followed by Gemini (Google) and Copilot (Microsoft).” This covers off-device AI, where your data is processed elsewhere.

This has real implications for the accelerating integration of Gemini and Gmail. Whether AI-powered relevancy searching or AI-powered smart replies, the warning is the same. This AI data processing takes place on Google’s servers not your device.

These Gmail AI upgrades are take it or leave it offers and come with clear privacy implications. Your data is being accessed and processed outside your control. And while you can control AI settings within your account, including what is stored and what is used for training, that offline processing is how it all works.

Contrast this with the Android privacy backlash this week when it seemed Gemini would be given unfettered access to Messages and WhatsApp on phones. Google clarified this to assure users that this is about giving users more control over their data, not less. The same reassurances cannot be made for Gmail’s own AI upgrades.

This is why Gmail’s AI upgrades do not work with Gmail’s quasi end-to-end encryption, because Google cannot see those emails to process them. Love it or hate it, that’s different to WhatsApp’s own AI upgrades which promise user data is not processed outside their own security enclave and is never visible to anyone else.

Meta fares worse than Google in this new report, and I have approached both Meta and Google for comment. But Meta does not control an OS nor does it control the platforms that run billions of lives. The hurdle for Google, Microsoft and Apple is higher, as AI is built into the platforms that we trust with all our most sensitive and personal data.

We have seen multiple privacy and security warnings for Google users as these AI upgrades are confirmed, including giving AI access to entire inboxes and even Drive. This is all undoubtedly useful and exciting and new. But it’s critical users understand the privacy risks and available opt-outs before they dive in and it’s too late.

“As these sophisticated models become increasingly integrated into daily workflows,” Incogni says, “the potential for unauthorized data sharing, misuse, and personal data exposure has surged faster than privacy watchdogs or assessments can keep up with.”

It’s decision time for Gmail users — where should you draw the line.