Latest AI News

Researchers at Switzerland's Paul Scherrer Institute, led by Giovanni Pizzi, developed XtalPaint, an open-source AI model that reconstructs missing atomic positions in crystal structures with a 97 percent success rate. Adapted from Microsoft's MatterGen and image inpainting techniques, XtalPaint applies diffusion noise only to unknown atomic positions, leaving known atoms intact. It restored correct hydrogen coordinates in 87 percent of tests and found more stable configurations in another 10 percent, potentially unlocking thousands of materials for battery and hydrogen storage research.

OpenAI has acquired NextSlide, a presentation startup whose product converts prompts, notes, documents, and research into polished, editable presentations. NextSlide founder Ahmed Beshry, previously a co-founder at Instacart-acquired Caper AI, announced the team is now working on ChatGPT. Beshry noted the announcement came "a few months late," as the deal closed earlier in 2026. Financial terms were not disclosed. The acquisition aims to expand ChatGPT's creation tools and make visual communication more accessible.

xAI has launched Imagine Image 2.0, now available as the Quality Mode on grok.com/imagine and its iOS and Android apps. The model features precise editing tools including a magic wand for region-specific edits, segmentation, background removal, and multi-reference editing supporting up to five input images. It also offers smart resize for any aspect ratio and pre-built templates for headshots, product shots, and game assets. xAI claims Image 2.0 ranks second globally in both text-to-image generation and image editing on Arena leaderboards.

Claude Code now supports cross-session messaging on macOS and Linux, allowing AI agent sessions to communicate with each other directly. Instead of re-explaining context manually, users can instruct Claude to send a summary to another session, which picks it up mid-task. The feature works bidirectionally, letting sessions ask each other questions and receive answers. Claude can also message other sessions autonomously, such as when a change it makes affects another session's work. Update Claude Code to access the feature.

Anthropic is making auto mode the default in Claude Code for Pro, Max, and Team plans starting August 14. Auto mode routes each tool call through a classifier that blocks irreversible or destructive commands, rather than relying on user approval prompts. In a controlled study with 1,053 paid testers, humans caught only 13.6% of dangerous commands while auto mode caught 89%. Teams using auto mode ship roughly 25% more pull requests. Enterprise and API users remain opt-in for now.

OpenAI has determined that its upcoming Astra model cannot rule out reaching the Critical cybersecurity threshold under its Preparedness Framework, meaning the model may be capable of identifying and developing functional zero-day exploits in hardened real-world systems without human intervention. In response, OpenAI is implementing isolated testing environments, enhanced model weight protections, universal monitoring for risky actions, and pausing internal Astra activities that don't meet new security controls. The company will also work with government agencies and AI safety organizations to test the model's capabilities.

Anthropic has updated the safety classifiers for Claude Fable 5, reducing biology-related fallbacks by approximately 85%. Previously, Fable 5 routed biology queries to the less capable Opus 5 model due to broad safeguards launched alongside the model. The refined classifier now better distinguishes benign requests—like interpreting lab results or learning about symptoms—from dual-use risks. Dual-use topics including virology, toxicology, and molecular design remain blocked. Anthropic plans trusted access pathways for professional biology researchers and drug developers.

Alibaba has launched Wan3.0 in public beta, a video generation model that doubles the maximum clip length of its predecessor to 30 seconds. Unlike most AI video tools capped at 15 seconds, Wan3.0 accepts multimodal inputs including text, images, video, audio, PDFs, and PowerPoint files, enabling conversion of static documents into dynamic video. The model features high-precision visual continuity, synchronized facial micro-expressions, and an intelligent duration recommendation tool. Users can access it via Alibaba Cloud's Model Studio and Qwen Cloud platforms.

Perplexity has launched Computer for Builders, a platform targeting solo founders and small engineering teams that orchestrates 15+ frontier AI models to automate the full software development lifecycle. Built on its existing Perplexity Computer product, it integrates with GitHub, Datadog, Stripe, Supabase, and Slack to write and deploy code, monitor production, track payments, flag fraud, and send automated weekly growth reports to Slack every Monday. It is available to Pro and Max subscribers.

OpenAI and former Apple designer Jony Ive are developing a hockey puck-sized, doughnut-shaped AI smart speaker expected to launch in 2027 for over $300, according to Bloomberg's Mark Gurman. The battery-powered device has no display and features moving parts that animate when responding, plus lights, a camera, and sensors. Designed to be carried around the home one-handed, it will run ChatGPT voice mode with more advanced models for humanlike interaction and may be the first in a planned family of OpenAI devices.

OpenAI's new country-by-country Signals dataset reveals ChatGPT is shifting from question-answering to task completion, with users at work more than twice as likely to use it for producing outputs like coding, editing, or analysis. Multimedia is the fastest-growing use case globally at 7.8% of messages, exceeding 10% in Brazil and Colombia following the April 2026 ChatGPT Images 2.0 release. Adoption gaps are narrowing, with Peru, Uruguay, and Costa Rica rising fastest in per-capita rankings, and users over 35 now account for a 5% higher share of messages year over year.

OpenAI has updated GPT-5.6 Sol in ChatGPT for Plus and Pro users, delivering more focused answers, fewer factual errors, and a new slider to control how much reasoning the model applies per response. Internally, factual errors dropped 68% compared to GPT-5.5 Instant. Free users get GPT-5.6 Luna as their new default model with unlimited text chats, plus a Think button for harder questions requiring deeper reasoning. Limits on file uploads and images still apply.

Google DeepMind's WeatherNext AI model has achieved a breakthrough in cyclone forecasting, published in Nature, delivering more than a full day of additional lead time for predicting a storm's track, intensity, and wind structure. Three-day forecasts now match what prior models achieved in two days, representing roughly a decade of meteorological progress. The model helped the National Hurricane Center forecast Hurricane Melissa's 2025 Jamaica landfall. DeepMind is now open sourcing WeatherNext 2 and WeatherNext Cyclones model weights on GitHub.

Anthropic and Millennium, one of the world's largest alternative investment firms, are co-developing a Claude-powered digital risk analyst to work alongside human risk managers. The AI teammate surfaces risk insights and forms opinions on exposure across asset classes using Millennium's proprietary data. Built in Millennium's internal AI lab, it retains reasoning over time, logs its actions in sandboxed environments, and requires human approval before decisions are enacted. Millennium already uses Claude and Claude Code across its plus investment teams.

Anthropic has launched self-hosted Claude Code environments in public beta for organizations on Team and Enterprise plans. Teams can now run Claude Code sessions on their own infrastructure, enabling access to internal services, databases, and registries without public internet exposure. Key benefits include customizable environments with pre-installed SDKs and compilers, and compliance controls keeping source code and build artifacts on-premises. Conversation data is still sent to Anthropic for inference. The feature is off by default and requires dedicated engineering staff for setup and maintenance.

Cloudflare has launched Kitesurf, an agent-first browser built entirely on Cloudflare Workers and designed specifically for AI automation. Unlike Chromium, which was built for humans, Kitesurf runs in V8 isolates using Rust and WebAssembly, making it significantly more efficient in CPU and memory for agentic tasks like screenshots and HTML extraction. It supports Chrome DevTools Protocol, meaning Puppeteer and Playwright work out of the box. Kitesurf is available for free in beta through Cloudflare's Browser Run product.

Google Maps has expanded its AI-powered Ask Maps feature with several major updates, including agentic food ordering through partners like Square, Toast, and soon Uber Eats, real-time transit widgets with minute-by-minute delay updates, and Personal Intelligence that connects to Gmail to factor in reservations and flights. Users can also search hotels and events with natural language, resume past conversations, and submit conversational map contributions. Ask Maps is now available in over 150 countries, with new features rolling out first in the U.S.

ElevenLabs has launched its Dubbing v2 API, allowing developers to embed AI dubbing directly into products and workflows. The new model preserves tone, emotion, and delivery across 90+ languages by conditioning on the original audio performance. Sync-aware translation aligns speech starts and stops automatically. Dubbing v2 also captures regional accents with sharper locale accuracy, distinguishing variants like Castilian and Latin American Spanish, and handles background music and multi-speaker scenes. Enterprises can supply custom transcripts or edit segments and regenerate selectively.

Suno has published its AI music principles alongside new transparency and anti-abuse tools. The music generation platform is rolling out audio watermarking and fingerprinting technology to help distribution platforms detect fraud and misuse, and will soon introduce a downloads policy to limit mass streaming distribution while preserving personal and professional use. Suno also runs Spark, a grants and mentorship program for unsigned artists, and works with Audible Magic and Musixmatch to screen uploads for unauthorized content.

Meta has launched Muse Code in beta, a terminal coding agent powered by Muse Spark 1.2, a coding-focused model update. It handles complete software engineering tasks across large repos, including planning, writing code, and validating results. Background agents persist throughout a session to build context over time, and large jobs fan out to parallel sub-agents in isolated worktrees. Every action is logged locally for crash recovery. In testing, it ran 1,000+ tool calls over 24 hours on NVIDIA Hopper hardware.

Demis Hassabis is stepping down as Google DeepMind CEO to become chair of Google DeepMind and chief scientist at Alphabet, with Koray Kavukcuoglu, formerly DeepMind's CTO, taking over as SVP of DeepMind reporting to Sundar Pichai. Hassabis will continue leading Isomorphic Labs, Google's AI drug development unit, citing a focus on curing diseases like cancer. Separately, chief scientist Jeff Dean and Google Fellow Sanjay Ghemawat are leaving to found Discovery Loop, an AI-focused public benefit corporation with Google as a founding investor.

Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, and Quoc Le — longtime Google collaborators with 14 to 30 years of shared history — have founded Discovery Loop, a Public Benefit Corporation aiming to automate machine learning, science, and engineering research. The startup will initially focus on automating large-scale ML experimentation, with ambitions to address NAE Grand Challenge problems. Radical Ventures and Khosla Ventures are leading the seed round, with Lightspeed, Kleiner Perkins, Doerr Capital, and Alphabet also participating.

Google DeepMind is reshuffling its leadership as Demis Hassabis transitions to Chair of Google DeepMind and Chief Scientist of Alphabet, stepping back from day-to-day operations to focus on AGI strategy. Koray Kavukcuoglu, a year DeepMind veteran who helped develop WaveNet and DQN, becomes SVP of Google DeepMind, overseeing Gemini model development and frontier AI research. The Gemini app has surpassed 950 million monthly users, and Gemma models exceeded 900 million downloads. Jeff Dean is also departing after 27 years to launch an independent public benefit corporation.

Reddit announced plans to modernize its infrastructure and moderation tools to support its 130 million daily users across 100,000+ communities. Key changes include expanding Rules Hub, an LLM-powered moderation suite tested in 700 communities that can eventually replace Automod, improving community discovery for new users, restricting public API access in favor of its Developer Platform backed by a $1 million migration program, and limiting Old Reddit access to combat scraping and abuse.

AI startup Hark, founded by serial entrepreneur Brett Adcock, launched Handoff, a web agent that autonomously completes tasks like ordering food on DoorDash or booking flights on United and Delta. Hark claims Handoff scored 97.7 on the Online-Mind2Web benchmark, topping OpenAI's GPT 5.4 at 92.8 and Anthropic's Claude Opus 4.8 at 84.1, at roughly one-tenth the token cost of rivals. Caveats exist: comparisons exclude newer frontier models like GPT-5.6 and Opus 5, and two of three benchmarks were run in Hark's own harness.

Anthropic is building a custom chip design team to make its Claude AI models run faster and more efficiently, confirmed to TechCrunch after Business Insider first reported the news. The company plans to co-design hardware and models together, and was reportedly scouting Samsung as a manufacturing partner. Anthropic already has computing deals with AWS, Google, Nvidia, and AMD, but rising Claude demand is pushing the company toward its own silicon, following OpenAI's Broadcom-built Jalapeño inference chip and Meta's MTIA accelerators.

The White House will not publicly release its new framework for evaluating advanced AI models, three sources told Axios. The voluntary framework, outlined in a June executive order, governs how AI developers work with the government to assess models before release, including up to 30 days of early government access. Only companies invited to staff-level meetings know its contents, leaving researchers, policymakers, and U.S. allies excluded. Nvidia participated in the meetings, and the benchmarking process for assessing cyber capabilities will be classified.

Black Forest Labs has launched FLUX 3 Video, a frontier multimodal model now generally available via the BFL API and select partners. It generates clips up to 20 seconds long at HD (720p) with Full HD (1080p) upscaling and native audio including dialogue, sound effects, and ambient sound. Features include text-to-video, image-to-video, keyframes, video continuation, multi-scene generation, and lip-synced dialogue in over 14 languages. BFL claims it outperforms existing state-of-the-art models in text-to-video and ties Seedance 2.0 in image-to-video.

OpenAI has launched three new education plugins for ChatGPT Work and Codex, targeting K–12 teachers, college educators, and college students. Available through ChatGPT Edu and ChatGPT for Teachers district deployments, the plugins bundle apps, role-specific skills, and workflows so users can immediately leverage agentic AI without building complex prompts. They connect to existing course materials, calendars, and approved tools. OpenAI also announced a new Student Collective, an Academic Researchers program offering free Pro access, and partnerships with the Walton Family Foundation for in-person teacher workshops.

Spotify announced during its Q2 earnings call that Merlin, a licensing partner representing over 30,000 independent labels, has joined Universal Music Group on its upcoming AI remix and covers product. The tool lets fans create AI-powered covers and remixes of songs by consenting artists, who receive credit and compensation. Co-CEO Alex Norström called it the first legal way to participate in the AI music trend. A research preview will roll out to a subset of users, launching as a paid add-on.

NVIDIA has released Alpamayo 2 Super, a frontier open reasoning model for robotaxis and autonomous vehicles, now available for commercial use under the OpenMDW-1.1 permissive license. Built on NVIDIA Cosmos 3 Super Reasoner and post-trained with reinforcement learning, it ranks first on the LingoQA autonomous driving benchmark, outperforming GPT-4o by 23.2 points. The billion-parameter model handles trajectory planning, chain-of-causation reasoning traces, and auto-labeling, and has surpassed 500,000 downloads on Hugging Face.

Meta, Anthropic, Google, and OpenAI staff are meeting with Trump White House advisers on Tuesday to discuss voluntary safety testing for advanced AI models, following disclosures that both OpenAI and Anthropic tools breached systems of other companies. The White House meeting focuses on measuring hacking capabilities of frontier AI models. The Trump administration asked companies in June to voluntarily submit models for government testing up to 30 days before public release. Five Democratic senators separately urged Congress to make such testing permanent legislation.

Autodesk Flow Studio has launched a 3D Editor and Canvas tool aimed at giving filmmakers precise directorial control over AI-generated video. Unlike most AI video tools that work in 2D, Flow Studio lets creators build and stage scenes in 3D, positioning characters, cameras, and performances from video sources as editable 3D animation. A node-based Canvas then renders and refines the final look using leading AI models. The platform builds on existing tools including AI MoCap, Camera Track, Animation, and Wonder 3D.

OpenAI has publicly pushed back against Apple's trade secrets lawsuit, releasing iMessages and emails it says undermine Apple's claims. The messages show Apple employees contacted former staffer Chang Liu after his January 2026 departure to ask for help locating files and technical information. OpenAI also revealed Apple's outside counsel emailed the wrong person by confusing two Asian last names, and that Apple's claim of a discussion with OpenAI General Counsel Che Chang never occurred. OpenAI says Apple told them issues were resolved, then filed suit five months later.

The EU's AI Act transparency obligations took effect August 2nd, requiring companies to disclose when users interact with AI and label AI-generated or manipulated content including deepfakes. Providers must notify users when dealing with AI instead of humans and embed machine-readable marks in synthetic media. Deployers must label realistic AI-generated audio, images, and video. The European Commission released optional disclosure icons to standardize labeling. Non-compliance risks fines up to €15 million or 3 percent of global annual turnover. Existing AI systems have until December 2nd to comply.

Alibaba has launched QwenWork, an all-in-one workplace AI agent platform now in public beta in China. Built on capabilities from its existing Qoderwork, Mulerun, and Wukong platforms, QwenWork unifies desktop, cloud, and enterprise collaboration agents in a single tool. It will integrate deeply with DingTalk, Alibaba's enterprise platform serving over 20 million organizations. Features include multimodal generation, web app creation, and autonomous agents. Users can access Alibaba's flagship Qwen3.8 Max model starting August 3.

Alibaba has launched Qwen3.Max, its largest and most capable flagship AI model to date, featuring 2.4 trillion total parameters with only 95 billion activated via a Sparse Mixture-of-Experts architecture. The multimodal model supports a 1 million token context window and ranks fifth in Text Arena and second in Vision Arena. It is available via API on Alibaba Cloud Model Studio, with model weights releasing next week. In testing, it autonomously ran a day software engineering project and outperformed humans in a multimodal dialogue challenge.

Genspark has open-sourced GenOffice, a free, ad-free AI-powered office suite for PC and Mac that includes Docs, Sheets, Slides, and PDF editing. Built by one engineer in one week using $10,000 in tokens, the Alpha integrates Genspark Super Agent to handle research, data analysis, and document creation. Users can shape the product by submitting feedback via GenTeam's group chat, with early contributors earning 1,000+ Genspark credits. The project is available on GitHub now.