Google upgrades Android with homework help and scam detection

Imagine picking up a call from someone claiming to be your bank, asking you to transfer funds immediately, and your phone screen flashes a warning that you are likely being scammed.

At I/O on May 14, Google unveiled a major push to wire AI deeper into Android, anchored by an upgraded on-device model and better system-level context awareness.

Homework help and a smarter screen overlay

First up is Circle to Search, which is morphing into a personal tutor. Powered by Google’s new LearnLM effort, students can now circle a stubborn physics or math word problem and get step-by-step instructions rather than just a dry final answer. Later this year, it will handle complex formulas, diagrams, and graphs. Circle to Search is already on over 100 million devices, with Google aiming to double that by the end of the year.

Meanwhile, Gemini is getting much better at recognizing what is happening on your screen. An updated overlay lets you ask questions about the YouTube video you are watching or use "Ask this PDF" if you are on Gemini Advanced. You can also drag and drop generated images straight into Gmail or Google Messages without app-hopping.

Gemini Nano gets eyes and ears on-device

Google is bringing full multimodality to Gemini Nano, its on-device foundation model, starting with Pixel devices later this year. That means the phone can process sights, sounds, and spoken language directly on silicon.

One big beneficiary is TalkBack. Blind and low-vision users encounter an average of 90 unlabeled images daily. Gemini Nano will generate vivid descriptions of clothing cuts or family photos locally, without lag and without requiring an internet connection.

Google is also testing a real-time scam detection feature for phone calls. With global fraud losses topping $1 trillion over a recent 12-month stretch, Gemini Nano will listen for conversation red flags, like callers demanding gift card payments or bank PINs, and flash an alert. Google emphasized this will be an opt-in feature running entirely on-device to keep audio private.

Why it matters

AI is finally moving out of isolated chatbots and into the plumbing of the OS. For everyday users, real-time scam alerts and homework breakdowns require zero prompt engineering; they just kick in when needed. If scam detection saves non-tech-savvy relatives from wire fraud, that is infinitely more valuable than generating generic emails.

Strategically, Google is leveraging its mobile home turf against rivals like OpenAI. Features like listening to live phone calls can only be built by the platform owner, and running them on-device sidesteps catastrophic privacy backlash.

That said, the execution threshold is razor-thin. If Gemini Nano falsely flags a frantic call from a family member asking for money as a scam, users will immediately toggle it off. Fine-tuning that alert sensitivity without cloud verification will be Google’s biggest engineering hurdle.