AI moved closer to the device, workflow and boardroom this week, with OpenAI, Google, Microsoft, Anthropic, Nvidia and Zoom promoting better assistants and on-device tools, while security researchers found new ways to sidestep those same systems. Add in major Android and web server flaws, AI-enabled scams, and SpaceX’s record IPO plans, and the message was clear: The next era of technology is fast approaching, with productivity gains and a suspiciously high number of flashing warning lights.
Important message
AI assistants and on-device intelligence are taking center stage
OpenAI introduced Dreaming V3, a new memory system for ChatGPT that integrates past chat details into user profiles. The update doubles storage for Plus and Pro users and adds privacy-oriented features like ad hoc chats. While improving system personalization, it also raises new privacy concerns as ChatGPT surpasses one billion monthly users.
Google introduced DeepMind Gemma 4 12B, a 12 billion parameter multimodal AI model that runs entirely on devices with 16GB of RAM. Capable of processing text, images, audio and video offline, the Gemma 4 12B supports a 256K context window and reduces reliance on cloud GPUs, providing developers with privacy and cost benefits.
Microsoft released Scout, an AI assistant embedded in Teams, Outlook, and across Windows. Built on the OpenClaw framework, Scout automates planning, communication, and task management as part of Microsoft’s shift toward an agent-first workflow and consumption-based AI model.
Microsoft further revealed its AI ecosystem, unveiling Project Solara, an Android-based OS designed for AI agent-driven devices. The platform supports adaptive interfaces and enterprise management tools, with pilot projects already being implemented in the retail and healthcare sectors.
AI Hardware and Cloud Partnership Expands
Nvidia and Microsoft have teamed up to launch the RTX Spark, an Arm-based superchip that combines a 20-core Grace CPU with a Blackwell RTX GPU and up to 128 GB of integrated memory. The chip powers the new Surface Laptop Ultra and promises a petaflop of on-device AI performance, with Spark-powered PCs expected this fall.
Pinterest signs $4 billion cloud and AI deal with Amazon Web Services through 2031 The deal leverages the AWS Trainium and Graviton chips and includes migration to Kubernetes for improved scalability and performance, increasing both companies’ stakes.
Insider Intel
AI management and security focus
In Build 2026, Microsoft introduced Microsoft Execution Containers (MXC), an OS-level sandbox that controls AI agents’ access to files, networks, and UI elements. This feature improves enterprise confidence by enabling real-time control and management through Microsoft’s security suite.
Anthropic Cloud introduced Opus 4.8, an AI model that emphasizes fairness and transparency. The update introduces a dynamic workflow that supports 1,000 subclasses, a faster and cheaper mode, and improved code defect detection. The company’s valuation is now higher than that of OpenAI because it follows the most trusted enterprise AI.
AI-Free Search Gains Speed
Following a backlash to Google’s AI-powered search results, DuckDuckGo saw a 70% jump in US iPhone installs and registered search traffic. The privacy-focused engine also introduced Chrome and Firefox extensions to protect AI-free search, while rivals like Brave and Bing experienced parallel growth from AI-wary users.
AI productivity tools are expanding
Zoom introduced Zoommate, an AI workspace that turns meeting discussions into actionable tasks and documents on platforms like Salesforce, Jira, and Slack. The new AI productivity suite allows users to create documents and slides directly within Zoom, seamlessly integrating with competing collaboration tools.
Safety warnings
Critical Vulnerabilities and Exploits
A zero-day flaw in GitHub’s browser-based VS Code allowed attackers to steal OAuth tokens with a single click, exposing private repositories. Researcher Ammar Asghar exposed the problem, and Microsoft advised users to provide mitigations, clear cookies, and review access tokens.
Researchers have discovered a new HTTP/2 “bomb” exploit capable of crashing major web servers – including NGINX, Apache, IIS, Envoy and Cloudflare Pingora – by abusing HPACK compression and flow control. Patches are available for NGINX, Apache, and Envoy, while IIS and Pingora are vulnerable.
Google released its June 2026 Android update, which fixed more than 120 other vulnerabilities, including a heavily exploited privilege escalation flaw and a remote zero-click exploit. Pixel devices were the first to receive the patch, with other vendors expected to follow.
AI and platform security risks
A vulnerability called fake context alignment allowed crafted notifications from apps like WhatsApp and Slack to manipulate Google Gemini on Android devices. Google last November pinned the issue as server-side and recommends turning back Gemini’s notification access for added security.
ChatGPhish, a new instant injection exploit, embeds malicious code into web pages. When ChatGPT summarizes such pages, it displays phishing links, fake alerts and tracking pixels. OpenAI has been notified, but has yet to issue a fix.
A flaw in Meta’s AI support bot allowed attackers to bypass two-factor authentication and reset Instagram passwords and hijack accounts. The exploit affected several high-level users before Meta deployed a fix.
Browser and device security
Google Chrome introduced Device Bound Session Credentials (DBSC) which binds session cookies to the device’s hardware security module. This feature prevents cookie theft attacks and is enabled by default on Windows and macOS, with administrative enforcement options.
Android adds Deepfake call detection to combat AI voice fraud. The feature, first rolling out on Pixel devices, uses RCS handshakes to verify callers and alerts users to possible spoofing or AI-generated voices.
Data breaches and fraud
Charter Communications confirmed a breach by the ShinyHunters group that exposed data from 13 million Spectrum customer accounts. The attack, fueled by a voice-phishing scam, compromised both customer and employee information.
The FBI reported nearly $900 million in AI-related fraud losses last year, leading to a rise in voice-cloning cons. Criminals use short audio clips to impersonate victims’ relatives or colleagues to trigger fraudulent transactions. Experts urge users to verify calls and limit public audio sharing.
Industrial vibrations
Major market moves and IPOs
SpaceX announced plans for a record $75 billion IPO, setting a $135 share price for its June 12 Nasdaq debut. The offering targets a valuation of $1.77 trillion — breaking Saudi Aramco’s record — and Elon Musk provides 82% of the voting power, fueling speculation about a possible Tesla merger.
Policy and Regulation
President Trump signed an executive order establishing a voluntary AI verification process that allows the NSA to review frontier AI models for 30 days before launch. The move aims to identify cyber risks, but has drawn mixed reactions from industry leaders and security advocates. OpenAI has called for mandatory federal oversight instead.
Corporate acquisitions and product changes
Asana acquired StackAI, a no-code agent builder, for $75 million to bolster its AI-driven workflow automation. The acquisition enhances Asana’s execution tools and positions it more competitively against monday.com and ServiceNow.
Meta is developing an AI pendant and new smart glasses as part of a plan to sell 10 million wearable devices by late 2026. Meta AI processing calls for recording conversations in the pendant, raising privacy concerns, while Apple’s competing AI glasses are said to be delayed until 2027.
Disney plans to roll out the Hulu app by the end of the year as part of Project Gemini, which plans to consolidate streaming operations under Disney+ to streamline content delivery and improve ad targeting, a leaked roadmap revealed.
IIf you’d like to see more from our newsletter, check out Daily Tech Insider Archive.
#upgrades #security #flaws #SpaceXs #record #IPO #define #TechTech #Republic #Week