Google AI Updates: May 13, 2026

1. DeepMind Reimagines the Mouse Pointer as a Gemini Surface

Google. Google DeepMind published the research framing behind its new “AI pointer,” a cursor that captures visual and semantic context from whatever it hovers over so users can issue shorthand requests like “this” or “that” without writing long prompts. The post lays out four design principles (maintain the flow, show and tell, the power of “this” and “that”, and turn pixels into actionable entities) and points to two interactive demos in Google AI Studio, one for image editing and one for map-based location finding. The interesting engineering bet here is that Gemini is being wired to treat pixels under the cursor as structured entities rather than screenshots, which is a meaningful shift from screen-grab-then-reason approaches. Source

2. Magic Pointer Ships on Googlebook and Lands in Gemini for Chrome

Google. The productized version of the AI pointer arrived as “Magic Pointer” on the new Googlebook line, with the same technology rolling into Gemini in Chrome so users can point at a section of a webpage and ask Gemini to compare products, visualize furniture in a room photo, or explain a chart without copying anything into a chat box. Google framed it as the end of “AI window detours,” where the assistant lives where the user already is rather than in a sidebar. Practitioners building agentic UIs should pay attention to how the pointer state and target region are passed into the model context. Source

3. Gemini in Chrome for Android Adds Auto Browse

Google. Chrome for Android is getting a full Gemini bottom-sheet next to the address bar, with the same feature set as desktop and iOS plus an Auto Browse mode that lets Gemini scroll, tap, and type to complete tasks like booking SpotHero parking or reordering on Chewy. The integration ties into Gmail, Calendar, Keep, and Google’s Nano Banana infographic tool, and ships at the end of June for US users on Android 12+ devices with 4GB of RAM, with Auto Browse gated to AI Pro and AI Ultra subscribers. Sensitive actions like purchases still require manual confirmation. Source

4. Android Show I/O Edition Repositions Android as an “Intelligence System”

Google. At The Android Show on May 12, Google bundled its mobile AI work under a new umbrella called Gemini Intelligence, with features spanning proactive task automation, AI-generated custom widgets, Magic Cue context fusion across Gmail/Calendar/Messages for Android Auto replies, and a new speech-to-text mode called Rambler that strips filler words and handles mid-sentence language switching without storing audio. Gemini in Chrome for Android is being built on Gemini 3.1 under the hood. Google also confirmed that Googlebooks, the new Android-based laptop line, will launch this fall optimized for Gemini Intelligence. Source

5. Google Threat Intel Flags First AI-Discovered Zero-Day Attempt

Google. Google’s Threat Intelligence Group published findings on what it believes is the first observed case of a cybercrime group using AI to discover and weaponize a previously unknown vulnerability, in this case a 2FA bypass in a Python script for an open-source sysadmin tool. Google attributes the attempt to AI based on telltale artifacts (overly explanatory code comments, fabricated severity ratings, characteristic AI-generated Python structure) and says it blocked the planned “mass exploitation event” before deployment and disclosed to the vendor. Analysts cited in the report note that North Korean APT45 and a Gemini-abusing Android malware family called PromptSpy are already operational. Source