Artificial intelligence is not your best friend, but avoiding it on Android is hard. Your opt-in is the act of , unless you get a nerfed or de-Googled device.
Google Assistant and Gemini are embedded in the operating system, and most manufacturers layer their solutions on top of them.
There's little visibility into what AI does in the background. Here are seven reasons to be concerned.
On-device doesn't mean no-logging
Your favorite AI may keep a record of what it sees later
The technology industry argues for on-device processing because your data never leaves the machine.
So, screen-aware AI doesn't introduce privacy risks beyond using the device. However, on-device processing merely describes where the computation happens in the moment.
Models may run locally on your phone or laptop at launch, and later send processed information back to company servers.
When Gemini Nano launched on the Google Pixel 8 Pro in 2023, Google's pitch was explicit: Sensitive data physically could not leave the phone.
But in 2025, the company introduced Private AI Compute to extend AI processing beyond on-device hardware capability.
It markets the service with "the same user security and privacy assurances of on-device processing."
Yet, it's like any other cloud-based platform that routes your data to remote servers for processing and returns the result to your device.
In the end, on-device processing is not a binding commitment. Companies can update their software at any time.
The Federal Trade Commission (FTC) handles such deceptive business practices. But they can't monitor every company's code minute by minute.
Your data may have already left your device before any intervention.
Your private texts aren't as private as you think
Encryption has one blind spot you're not seeing
End-to-end encryption (E2EE) is a widely appreciated perk on social media. It's comforting knowing your conversations are safe from anyone in between.
However, E2EE scrambles while in transit from your phone to the destination. The content is prey for compromised screen-reading AI when a device decrypts it, and it becomes plain text.
Microsoft users may remember the Recall feature launched with Copilot+ computers in May 2025. It took repeated screenshots for the company's AI to index.
It became a privacy nightmare because it captured messages, passwords, banking details, medical records, and more indiscriminately.
Security researchers found the database was poorly protected, and Microsoft made it opt-in after the backlash.
Your weaknesses are training data
Without it, AI can't provide strong answers
The US government forced Anthropic to pull its new models offline in June 2026. In its defense, Anthropic had spent months warning about its own AI.
It framed Claude Mythos Preview as too dangerous in the cybersecurity domain to release without serious restriction. So, it was comical that Amazon, one of the company's investors, proved the dangers existed.
Amazon used Fable 5 to demonstrate how a software vulnerability could be exploited. Although Fable isn't a local AI tool, its cybersecurity training isn't exclusive to it.
OpenAI and Google train their models similarly. In particular, Gemini has evolved from a chatbot into a deeply embedded Android feature.
It has documented permission to call logs, contacts, message history, and other system permissions. This permission has already been exploited.
In 2025, Noma Security discovered GeminiJack, where a poisoned Google Docs entry, calendar invite, or email could get Gemini to steal data without you touching anything.
Google patched the vulnerability. But it continued letting its AI access sensitive information whether you turned Gemini Apps Activity off or not.
The competition is getting your secrets for free
Training with human data is proof of unoriginal ideation
It's just business if your competitor outworks you and gets ahead. But it's possible you helped train the tools that made them successful.
Some AI providers reserve the right to use captured screen content, documents, and prompts to improve future models.
It means they want to train on your data, which is why skipping those long, boring terms and conditions may come back to haunt you.
Your unpublished work or internal numbers could train models for your competitor queries. Neither of you would ever know it happened.
In November 2025, plaintiffs alleged that Figma used customer design files to train its generative AI tools after years of assuring users of privacy.
The AI allowed competitors with accounts on the platform to query the leaked information, including rivals who never had access to the original works.
For context, Figma's clients reportedly include Alphabet, Microsoft, and Netflix, so the files in question weren't hobbyist sketches.
Identity theft is virtually real
People may pretend to be you
Accessibility Service has been part of Android since 2009. It was originally for screen readers and other assistive tools.
It can read everything on your screen, detect app changes, access your clipboard, and perform actions on your behalf.
Sadly, Android doesn't let you grant screen-reading access without those broader privileges. Attackers may exploit it.
PromptSpy, the first Android malware to weaponize generative AI, masquerades as a JPMorgan Chase app.
It sends on-screen content to Gemini, and uses the AI's instructions to stay hidden. Then it steals your lock screen PIN, records your screen, and gives attackers remote control of your device.
The ads never end
No matter how much you try to block them
Personalized ads are still ads, and most ads are annoying. Americans pick up their phones 186 times daily, and receive an average of 46 push notifications daily. Gen Z users receive 181 daily.
It exposes poor screen time management and how easily marketers can monetize your attention. AI turns it into a detailed trail, where traditional ads would normally guess.
It may even make suggestions based on your feelings. Regardless, suggestions come from context gathered over time.
AI is unsettling
It gives me goosebumps when I think of it
I like to read about technology's dark side, occasionally. I've found countless forum posts echoing people's concerns about privacy. They felt their phone was reading their minds.
Personally, I've thought of something, only to open TikTok or Instagram minutes later and see a related post. Those coincidences have happened too many times for me to wave them off.
Coupled with the long-running belief that phones eavesdrop through passive listening, AI isn't doing much to ease the concern.
If anything, it makes the feeling of someone watching you stronger.
Dial down the AI noise
AI doesn't have to be intrusive. Google has offered ways to reduce its reach into your data. You can uninstall the Gemini app entirely or turn off Gemini in Messages.
The options are scattered across multiple menus, so you'll have to disable smart features separately. If you use a Samsung Galaxy phone, Galaxy AI settings are in one menu.