Latest AI News

The accessible source details point to this update: X Launches Hosted MCP Server to Connect AI Apps With Its API. Because the full article could not be reliably extracted or rewritten, this TLDR stays conservative and is based on headline-level information plus limited source context from techcrunch.com.

Startup OASIS has unveiled the OASIS 1, a smart ring combining AI-powered whisper dictation with a built-in capacitive trackpad for cursor control and text editing. The ring uses WisprFlow's dictation technology to transcribe speech privately, letting users write without speaking aloud. A noise-isolating microphone and haptic feedback trackpad allow subtle finger gestures to navigate and correct text. The device offers up to 16 hours of battery life and is available to pre-order for $289, with first shipments expected around Christmas 2026.

Google is rolling out updates to Gemini Spark, including a new macOS app that lets users automate desktop tasks like sorting PDFs into folders or building budget spreadsheets from local invoices. Spark can also run multi-step tasks remotely from a phone. New app integrations include Canva, Dropbox, Instacart, OpenTable, and Zillow Rentals, plus Google Tasks and Keep. Custom MCP support is also launching. The macOS beta is available to Google AI Ultra subscribers in the US.

Anthropic announced that the U.S. Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5, two of its AI models. The company stated it will begin restoring user access the following day and plans to share a further update soon. Anthropic thanked users for their patience during the restriction period and acknowledged those who worked with the company throughout the process of getting the controls lifted.

The Trump administration plans to lift export restrictions on Anthropic's Fable 5 AI model as early as Tuesday evening, restoring general user access after weeks of industry disruption. Fable 5 is a consumer version of Anthropic's more powerful Mythos 5 model, built with added safeguards against cyberattack misuse. Export controls were first imposed June 12; Mythos restrictions were partially lifted June 26. The move reflects the administration's effort to balance cybersecurity concerns with AI competitiveness against China.

Runway has integrated Google's Gemini Omni Flash model into its AI video platform, giving users a new option for generating and editing video content. The feature supports multiple input types, including text prompts, images, and existing video clips, making it flexible for a range of creative workflows. Users can access Gemini Omni Flash directly through Runway's interface or by instructing Runway's built-in Agent to use Omni, expanding the platform's available model lineup for AI-driven video production.

Netflix's Wonka's The Golden Ticket, premiering September 23rd, uses an AI-generated voice of Gene Wilder for its voiceover. Netflix partnered with AI audio company ElevenLabs and obtained consent from Wilder's family. The reality show follows 12 golden ticket winners competing over one week, with a two-part finale on September 30th. ElevenLabs previously recreated the voices of Michael Caine and Stan Lee. The show continues Netflix's 2021 partnership with the Roald Dahl company.

Google has launched two new generative media models: Nano Banana 2 Lite, its fastest and most cost-efficient image model at $0.034 per 1,000 images with a second text-to-image latency, and Gemini Omni Flash, a video generation and conversational editing model priced at $0.10 per second of output. Both are available in Google AI Studio, the Gemini API, and Gemini Enterprise Agent Platform. Nano Banana 2 Lite is also rolling out to Search, Gemini app, NotebookLM, and Google Ads.

Anthropic's Claude Code team has published a guide defining agentic loops as agents repeating cycles of work until a stop condition is met. The guide covers four loop types: turn-based loops triggered by user prompts, goal-based loops using the /goal command with verifiable exit criteria, time-based loops using /loop and /schedule for recurring tasks, and proactive loops that run autonomously on events or schedules. The team recommends starting simple, using SKILL.md files for self-verification, and piloting before large runs to manage token usage.

Anthropic has launched Claude Science, an AI workbench for researchers, now in beta for Claude Pro, Max, Team, and Enterprise users on macOS and Linux. The app consolidates fragmented scientific tools into one environment, offering over 60 pre-configured skills and connectors for genomics, proteomics, single-cell analysis, and cheminformatics. It integrates with NVIDIA's BioNeMo Agent Toolkit and manages compute across local machines or HPC clusters. A reviewer agent checks citations and calculations automatically. Anthropic is also offering up to $30,000 in credits for select AI for Science projects.

OpenAI has launched GeneBench-Pro, a research-level benchmark testing AI agents on judgment-heavy computational biology tasks across 129 problems spanning 10 domains including statistical genetics, cancer genomics, and clinical diagnostics. Unlike traditional benchmarks, problems use synthetically generated data with known causal structures to enable deterministic grading. OpenAI's strongest model, GPT-5.6 Sol, achieved a 28.7% pass rate, up sharply from below 5% when GPT-5 was tested on the original GeneBench. Ten representative questions are open-sourced on Hugging Face.

OpenAI engineers investigating mysterious crashes in Rockset, a C++ data infrastructure service powering ChatGPT search and conversation plugins, discovered two unrelated bugs using an epidemiological approach. By building an automated pipeline with ChatGPT to analyze a year's worth of core dumps, they separated crashes into two distinct populations: one traced to silent CPU arithmetic errors on a single faulty Azure physical host, and another to an year-old race condition in the widely used open-source GNU libunwind library, which has now been patched.

OpenAI teased a new hardware device tied to its AI coding tool Codex, set to launch on July 15th in partnership with Work Louder, a company known for mechanical keyboards and macro pads. A video posted to X shows a square-shaped device with multiple buttons, captioned "Your favorite Codex shortcuts are getting an upgrade." The device resembles Work Louder's Creator Micro 2, which features 13 mechanical switches, a joystick, and a touch sensor for assignable shortcuts.

Google's Gemini 3.5 Pro is on track for a July launch after missing its June target, making it the only major frontier AI model set to release without U.S. government restrictions. Unlike Anthropic's Fable 5, which was forced offline June 12 after NSA red-team tests, and OpenAI's GPT-5.6, restricted due to a 96.7% cybersecurity benchmark score, Gemini 3.1 Pro scored 70.7% on Terminal-Bench 2.1 — well below the unofficial threshold. Gemini 3.5 Pro will feature a million-token context window and a Deep Think reasoning mode.

Anthropic has launched the Claude Apps Gateway, a new tool that connects Claude Code to Amazon Bedrock and Google Cloud. The gateway runs as a stateless container on your own infrastructure and provides enterprise features including corporate SSO login, centrally enforced policy controls, role-based access management, per-user cost attribution, and spend caps. The solution gives organizations centralized oversight and cost control when deploying Claude Code across teams on major cloud platforms.

Firefly Aerospace's Blue Ghost Mission 2, targeting late 2026, will be the first mission to run NVIDIA Jetson edge AI in lunar orbit. Firefly's Ocula imaging service, powered by Jetson modules built by Lawrence Livermore National Laboratory, will process ultraviolet and visible spectrum imagery onboard rather than downlinking raw data. This compresses analysis from weeks to near real time. Firefly's Elytra spacecraft will orbit the moon for five years running Ocula, while a lander descends to the far side carrying a radio telescope for UC Berkeley-led NASA research.

Google has launched its Take notes for me feature in Google Meet for AI Pro and Ultra subscribers, available now on web and mobile in select languages. Powered by Gemini, the tool transcribes conversations in real time, generates meeting summaries with key action items, and automatically saves notes to a Google Doc in Drive. After the meeting, users receive an email recap. Eligible Workspace business customers can also access it. All participants are notified when note-taking is enabled.

Google is expanding personalized image creation in its Gemini app to all eligible U.S. users for free. The feature uses Personal Intelligence to pull data from Gmail, Google Photos, YouTube, and Search with user permission. By connecting with Nano Banana and Google Photos, Gemini can generate tailored images from simple prompts like "design my dream house" or "create an illustration of me and my favorite things" using actual photos from your library. The feature is opt-in and adjustable in settings.

Anthropic's Claude models are now generally available in Microsoft Foundry, hosted on Azure. Claude Opus 4.8 and Claude Haiku 4.5 are accessible via the Messages API with features like prompt caching and extended thinking. The integration uses existing Azure identity, networking, and governance controls, with a single consolidated invoice and support for Microsoft Enterprise Agreement commitments. A US data zone option addresses data residency requirements. Anthropic operates inference and acts as data processor, with NVIDIA GB300 GPUs powering the deployment.

Cursor has launched a native iOS app in public beta, available to all paid plan subscribers. The app lets developers launch AI coding agents in the cloud or remotely control agents running on their local machine from their phone. Features include voice input, slash commands, Live Activities on the lock screen, push notifications, and the ability to review diffs and merge PRs directly from the app. Cursor is offering 75% off Composer 2.5 runs in the mobile app through July 5, 2026.

California Governor Gavin Newsom announced a partnership with Anthropic to provide Claude AI to state agencies and local governments at a 50% discount. The deal aims to strengthen cybersecurity across California's public sector. Newsom framed the agreement as a way to ensure all Californians benefit from responsible AI innovation, noting that the Golden State helped build Silicon Valley and should share in the advantages of its latest technologies.

xAI's Grok 4.5, built on a 1.5 trillion parameter V9 foundation model with Cursor data added during supplemental training, has entered private beta at SpaceX and Tesla. Early evaluations show performance close to or potentially exceeding Claude Opus. Reinforcement learning continues to improve the model, and the Grok Build harness improves daily. Elon Musk also announced that completely new models trained from scratch will be released by SpaceX on a monthly basis throughout 2025.

HP Inc. has announced a Frontier strategic partnership with OpenAI, scaling beyond successful pilots launched in February 2026. Early results included one engineer completing 122 pull requests across 43 projects in weeks, and a security team remediating software bugs in a day that would have taken a month. The partnership will deploy AI across customer support, partner portals serving 100,plus partners, device fleet management, cybersecurity, and software development, with OpenAI estimating roughly 82 hours per week of security-team capacity unlocked.

YouTube has added AI-powered comment moderation tools to YouTube Studio, available on desktop for all creators globally. The update introduces a new Search filter that finds comments based on broader meaning or topic rather than exact keywords. Creators can type phrases like "comments about my appearance" or "questions about my gear" to group related comments, or choose from suggested topics like "Excitement and enthusiasm" or "Negative feedback." A "Find similar comments" option also lets creators locate comments with similar meaning. The tools do not auto-remove anything.

Google Finance is exiting beta with major upgrades, including a new dedicated Android app. Users can track portfolios via a single dashboard showing performance and asset allocation, with setup via screenshots, CSV or PDF uploads, or plain-text descriptions. A new market intel feature lets users schedule custom AI-generated briefings, such as daily pre-market cryptocurrency updates. The Android app offers real-time data, a live news feed, an AI research tool, and AI-powered key moments explaining stock moves. An iOS app is planned for later in 2026.

Runway has launched Agent 2.0, a new tool built into its platform that lets users go from a simple prompt to fully realized marketing briefs and campaign assets. The system can also analyze performance data to help improve creative output and scale it across platforms, formats, and markets. Agent 2.0 represents Runway's push into AI-powered marketing workflows, streamlining the process from ideation to multi-platform campaign execution.

Mark Cuban argues that opposition to AI data centers is actually a proxy for broader anger over AI's economic disruption and wealth concentration. He urges OpenAI and other major LLM companies to launch community tours focused not on explaining AI benefits, but on helping towns facing job losses. Cuban also calls on AI leaders to directly engage artists and creative unions in LA and NYC, warning that buying politicians and celebrity endorsements will backfire. He says big LLMs have lost the PR battle by failing to put people first.

Microsoft has updated Copilot in Excel with new finance-focused features including skills, data connectors, and traceability tools. Skills let teams define repeatable workflows like DCF models or variance analyses using a SKILL.md file saved to OneDrive. New financial data connectors include CB Insights, Daloopa, FactSet, Morningstar, PitchBook, and S&P Global, joining existing LSEG and Moody's integrations. A Plan with Copilot feature outlines intended changes before execution, and edits are now attributed to Copilot in the Show Changes pane for full auditability.

Researchers at the University of Chicago Pritzker School of Molecular Engineering have developed a flexible AI skin patch that processes health data directly on the body in milliseconds, without relying on external servers. Built with organic electrochemical transistors printed onto flexible materials using UV-hardened polymer gel, the device can pack up to 10,000 transistors per square centimeter. In tests, it ran algorithms to map heart rhythms and detect ventricular fibrillation fast enough for real-time intervention, a task impossible with cloud-dependent wearables. The study was published in Nature Electronics.