★ Top Story
Google AI BlogGoogle's I/O 2026 keynote marked a decisive pivot from chatbots to autonomous AI agents, with Sundar Pichai framing the Gemini era as one where AI doesn't just answer questions but takes action on your behalf. The company unveiled a sweeping set of announcements spanning new models, redesigned search, agentic assistants, smart glasses, and developer tools. DeepMind CEO Demis Hassabis closed the event by suggesting we may be standing at the "foothills of the singularity."
Read the original →
The VergeGoogle has fundamentally reimagined its search experience, retiring the 25-year-old keyword-input paradigm in favor of a dynamic, AI-driven interface that fluidly blends AI Overviews, conversational AI Mode, and autonomous agents. The redesigned search box — the first major change to the interface in a quarter century — can now accept voice, images, and natural language, and route queries to specialized agents that act on results rather than just listing them. Publishers face further traffic erosion as Google increasingly answers queries without sending users to external sites.
Read more →
Google AI BlogGoogle released the Gemini 3.5 model family at I/O, headlined by Gemini 3.5 Flash — a fast, agent-optimized model capable of autonomously executing complex tasks and writing software from scratch. Alongside it, Gemini Omni is a new multimodal model that reasons across text, images, audio, and video to generate and edit video through natural conversation. Together, the models signal Google's strategic shift from building a better chatbot to building AI that can independently get things done.
Read more →
TechCrunch AIGoogle announced Gemini Spark at I/O 2026, a 24/7 agentic assistant built on Gemini's base models and the Google Antigravity agentic framework, with deep integration into Gmail and other Workspace apps. Unlike a traditional chatbot, Spark can proactively monitor topics, manage tasks, and take actions on a user's behalf without being explicitly prompted each time. The launch raises significant questions about data privacy, as the assistant requires broad access to personal communications and calendars to function.
Read more →
The VergeGoogle is bringing Gemini Live's conversational voice experience directly into Gmail with a new feature called Gmail Live, allowing users to speak naturally to find buried emails, summarize threads, or draft replies. The same voice-based AI prompting is also rolling out to Google Docs and Keep, letting users dictate notes and create drafts hands-free. The updates are part of a broader Workspace push to make Gemini the connective tissue across all of Google's productivity tools.
Read more →
TechCrunch AIGoogle is launching "information agents" — persistent background AI processes that track topics a user cares about and surface relevant updates without requiring a new search query. The agents represent a fundamental shift in how Google envisions search: less a reactive lookup tool and more a proactive intelligence layer woven into daily life. Users can configure agents to watch for news, price changes, research developments, or any other evolving topic.
Read more →
The VergeNearly every major announcement at Google I/O 2026 — from Gemini Spark to Gmail Live to Universal Cart — hinges on users granting Google's AI systems broad access to their personal data, communications, and browsing behavior. Google has framed this as a convenience trade-off, but critics note the company is asking for an unprecedented level of intimacy with users' digital lives at a time when AI systems still hallucinate and make errors. The tension between AI utility and data privacy is emerging as the defining challenge of the agentic AI era.
Read more →
The VergeGoogle unveiled Universal Cart at I/O, a cross-retailer shopping agent that lets users add products from any website into a single Google-managed cart, with Gemini able to complete purchases autonomously. The feature is designed to extend into YouTube and Gmail, turning passive content consumption into a seamless commerce experience. The announcement puts Google in direct competition with Amazon's checkout ecosystem and signals a major push into AI-driven commerce.
Read more →
TechCrunch AIGoogle revealed "audio glasses" at I/O — voice-commanded wearables that tap into Gemini and Google's app ecosystem, drawing a clear parallel to Meta's Ray-Ban smart glasses strategy. Separately, Google and Volvo announced that Gemini will gain access to external cameras on the upcoming EX60 SUV, enabling the AI to read parking signs, interpret surroundings, and assist drivers contextually. Both announcements illustrate Google's push to embed Gemini into physical-world hardware beyond the smartphone.
Read more →
TechCrunch AIGoogle expanded AI Studio at I/O to support native Android app generation, allowing anyone to describe an app idea in plain language and preview it instantly in an embedded Android emulator. A companion Android CLI tool enables AI coding agents like Claude Code and OpenAI Codex to build and iterate on Android apps directly from the command line. The move dramatically lowers the barrier to Android development and accelerates the "vibe coding" trend where AI does the heavy lifting.
Read more →
TechCrunch AIGoogle DeepMind announced that its Genie generative world model is now integrated with Street View data, enabling interactive simulations of real-world locations that can model weather changes, rare scenarios, and navigable environments. The technology has immediate applications in robotics training, gaming, and immersive travel experiences. It represents a significant step toward AI systems that can reason about and act within physically accurate representations of the real world.
Read more →
Ars TechnicaGoogle's SynthID watermarking technology, originally developed to embed imperceptible signals in AI-generated content, is gaining industry-wide adoption with OpenAI and NVIDIA among the new partners. Simultaneously, OpenAI announced it is joining the open C2PA provenance standard and embedding SynthID into its own image generation products to help users verify whether images are AI-generated. The coordinated move represents the most significant cross-industry effort yet to establish a reliable chain of custody for AI-generated media.
Read more →
TechCrunch AIA jury swiftly rejected Elon Musk's lawsuit against OpenAI, Sam Altman, Greg Brockman, and Microsoft, which alleged that Altman had fraudulently steered the nonprofit away from its founding mission. Evidence presented at trial undermined Musk's position by showing he had harbored similar ambitions to commercialize the organization, weakening his claim of being uniquely wronged. MIT Technology Review's coverage of the trial noted the verdict was consistent with what observers saw in the courtroom throughout proceedings.
Read more →
TechCrunch AIAndrej Karpathy, the influential AI researcher and OpenAI co-founder who previously founded the AI education startup Eureka Labs, has joined Anthropic to work on its pre-training team — the group responsible for the large-scale training runs that give Claude its foundational knowledge. The move is a significant talent acquisition for Anthropic as competition among frontier AI labs intensifies. Karpathy's deep expertise in neural network training and AI education makes him a high-profile addition at one of the most compute-intensive phases of model development.
Read more →
Ars TechnicaTwo independent AI-based research assistants demonstrated the ability to generate viable scientific hypotheses for drug-retargeting — identifying existing approved drugs that could treat different diseases — with one system going further to autonomously analyze supporting data. The results suggest AI is maturing from a literature-search aid into a genuine scientific collaborator capable of contributing to early-stage drug discovery workflows. Researchers caution that human expert validation remains essential, but the findings point toward a near-term future where AI meaningfully accelerates pharmaceutical research.
Read more →
TechCrunch AICybersecurity startup Ocean has raised $28 million to deploy an agentic AI platform that analyzes the full context of every incoming email — not just surface signals — to detect sophisticated fraud and impersonation attempts enabled by generative AI. The company was founded by a former Israeli military cybersecurity researcher who worked on Iron Dome-adjacent systems, bringing a defense-grade threat-modeling approach to enterprise email security. The funding round reflects growing enterprise urgency around AI-generated phishing, which has made traditional signature-based filters increasingly ineffective.
Read more →
Ars TechnicaSensitive credentials belonging to the Cybersecurity and Infrastructure Security Agency — including SSH keys and plaintext passwords — were discovered sitting exposed in a public GitHub repository, where they had been accessible since at least November 2025. The incident is a significant embarrassment for the agency responsible for advising the rest of the U.S. government on cybersecurity hygiene. It underscores the persistent and often mundane nature of credential-exposure vulnerabilities, even within organizations that exist specifically to prevent them.
Read more →No stories from this source today.