Apple is quietly working on a "ChatGPT-like search experience" to generate instant answers from the web, according to Bloomberg. Spearheaded by its internal AKI team, the answer engine could eventually power Siri, Safari, and Spotlight, or even launch as a standalone app. Why it matters: With its massive Google search deal facing antitrust pressure, Apple is preparing to take search into its own hands.
Cisco Talos researchers just revealed a technique called decomposition that tricks AI models into bypassing safety guardrails and repeating human-written training text verbatim. In testing, the method extracted exact sentences from dozens of New York Times articles, reopening fierce debates around copyright infringement and enterprise data leaks.
Google's Kaggle just launched Game Arena, a benchmarking platform designed to test model reasoning through strategic gameplay. From Aug. 5-7, eight flagship models—including OpenAI's o3, Gemini 2.5 Pro, and DeepSeek-R1—will compete in a live chess tournament with zero access to external chess engines like Stockfish. Why it matters: Unlike easily saturated static benchmarks, game theory provides a dynamic test of planning, memory, and logic under incomplete prompt conditions.
Cloudflare just published research showing Perplexity routinely bypasses Robots.txt blocks by disguising its crawlers as macOS Chrome browsers across tens of thousands of domains. Perplexity dismissed the report as a sales pitch and claimed the flagged bot isn't theirs, while Cloudflare booted the startup from its verified bot list.
Google just expanded NotebookLM to younger users, allowing teens aged 13+ and all Workspace for Education students to turn notes into interactive mind maps and a podcast-like Audio Overview. Google added stricter safety filters for minors and pledged never to train AI models on their notes or chats.
Alibaba just released Qwen-Image, an open-source 20B MMDiT foundation model built to tackle complex text rendering in both English and Chinese. It handles multi-line typography, paragraph layouts, and precision editing without garbling letters. It delivers native bilingual graphic design directly to the open-source image ecosystem.
OpenAI just shifted its core metrics for ChatGPT, moving away from screentime and clicks toward whether users actually solved their problems and moved on. The chatbot is rolling out gentle break reminders for long sessions, adding safeguards against emotional dependency with input from over 90 physicians, and refusing to make high-stakes life decisions for you.
Anthropic just introduced an early framework for developing safe and trustworthy AI agents as autonomous tools take off. The guide outlines five principles, including balancing human control with autonomy, keeping decision-making transparent, and safeguarding against prompt injection attacks. It is designed to establish emerging standards before agents take over high-stakes workflows.
Elon Musk's xAI just rolled out Grok Imagine to SuperGrok and X Premium+ subscribers on iOS. The tool whips up images in seconds and turns prompts into 15-second videos with native audio. True to Musk's ethos, it features a "spicy mode" that permits semi-nude content, though strict moderation still blocks overt explicit prompts and certain celebrity deepfakes.
Anthropic just partially blocked OpenAI's developer API access to Claude, accusing its rival's engineers of misusing the Claude Code tool ahead of the GPT-5 launch. Anthropic stated its terms strictly forbid reverse-engineering or training competing models, though benchmark access remains open. OpenAI called the cutoff disappointing, arguing that comparative safety evaluations are standard practice across the industry.
Microsoft just analyzed 200,000 anonymized Copilot conversations to map out the 40 jobs most vulnerable to generative AI. Occupations focused on processing and communicating information—including translators, writers, and customer support—sit at the top of the risk list, while physical trades remain largely untouched. Still, the researchers warn that high AI usage doesn't necessarily mean outright layoffs.
Brilliant Labs just unveiled the Halo smart glasses, a $299 rival to Meta's Ray-Bans featuring a peripheral microOLED display and bone-conduction speakers. Its Noa AI assistant uses a memory system dubbed Narrative to recall names and past conversation details when you meet people, along with an experimental voice-coding mode. It gives wearers a lightweight, heads-up second brain right on their face.
Google just launched Deep Think in the Gemini app for Google AI Ultra subscribers, introducing parallel thinking techniques that let the model explore multiple hypotheses simultaneously. Built to tackle complex coding and research problems, it hits Bronze-medal performance on internal 2025 IMO benchmarks while remaining fast enough for daily use. Subscribers get access via a toggle, though it comes with a fixed daily prompt limit.
Apple CEO Tim Cook just confirmed the company is "open" to M&A deals as it significantly ramps up AI spending and reallocates internal talent. While Apple Intelligence and an upgraded Siri face delays, reports suggest Apple may tap outside giants or pursue acquisitions like Perplexity to catch up.
Anthropic just took the crown from OpenAI as the top LLM provider for businesses, capturing 32% of enterprise usage compared to OpenAI's 25%, according to Menlo Ventures. Anthropic is especially dominating in coding with a 42% market share—more than double OpenAI's 21%—largely driven by Claude 3.5 and 3.7 Sonnet.
Uber Eats just rolled out generative AI features that draft menu descriptions, summarize diner feedback, and digitally touch up or re-plate low-quality food photos. To balance out overly polished AI images, the app is also paying diners in-app credits to upload their own raw food pictures when leaving reviews.
OpenAI just unveiled Stargate Norway, its first AI data center initiative in Europe. Partnering with Nscale and Aker, the project aims to deploy 100,000 NVIDIA GPUs by the end of 2026, running entirely on cheap local hydropower with direct-to-chip liquid cooling.
Z.ai just introduced its new flagship MoE models, GLM-4.5 and GLM-4.5-Air, prioritizing depth over width to enhance reasoning and autonomous agent capabilities. GLM-4.5 features 355B total parameters with only 32B active, launching with day-one support on SGLang alongside FP8 variants. Why it matters: it hits a 90.6% tool-calling success rate and outpaces Claude 4 Opus on web browsing, making it a formidable foundation for autonomous workflows.
Krea just released open weights for FLUX.1 Krea [dev], an image model built alongside Black Forest Labs designed to ditch the generic "AI look." Fully compatible with the FLUX.1-dev pipeline, it relies on opinionated post-training with fewer than 1M hand-curated samples, ditching waxy textures and artificial blur for authentic, photorealistic aesthetic control.
Manus just launched Wide Research, a new parallel processing feature designed to scale agent compute by up to 100x. Instead of using rigid pre-defined roles, the system orchestrates multiple fully capable Manus instances to research hundreds of items simultaneously. It is currently available to Pro users, with other tiers coming soon.
Meta CEO Mark Zuckerberg just claimed that glasses are the ultimate form factor for AI, warning that people without them will face a significant cognitive disadvantage. Because glasses let AI see and hear what you experience in real time, he plans to add displays to unlock even more value.
TikTok just introduced a new batch of safety features, including US-piloted Footnotes for crowdsourced fact-checking and tighter parental notifications. For creators, it rolled out Creator Care Mode, which uses AI to learn disliked comments and filter them out automatically.
Higgsfield AI just integrated MiniMax's Hailuo video model, unlocking unlimited free generations alongside a $20,000 creator challenge on Instagram. Creators can submit short AI-generated clips until August 17 to compete for top prizes. Why it matters: It completely removes credit barriers for creators eager to test-drive frontier AI video.
Anthropic just signed the CMS Health Tech Ecosystem pledge to help modernize healthcare data sharing across the United States. The company plans to use its Model Context Protocol to bridge fragmented record systems so patients and doctors can easily access information. Why it matters: Healthcare silos are notoriously difficult to break, and Anthropic is betting conversational AI can solve the interoperability bottleneck.
Fable just publicly launched Showrunner, an AI tool hyped as the Netflix of AI that turns text prompts into animated TV scenes and full episodes. Backed by fresh funding from Amazon's Alexa Fund, the service lets users direct story worlds or put themselves into shows, currently available for free via Discord.
Google DeepMind just launched AlphaEarth Foundations, a virtual satellite AI that merges optical, radar, and climate data into unified embeddings. Available in Google Earth Engine, it reduces storage needs by 16x and cuts error rates by 24% to help researchers track environmental changes.
Meta CEO Mark Zuckerberg just outlined his vision for "Personal Superintelligence," arguing that advanced AI should empower individuals rather than centrally automate all human work. Zuckerberg noted that AI systems are already beginning to self-improve, predicting that context-aware smart glasses will become our primary interface for these personal companions.
Google just agreed to sign the European Union’s General Purpose AI Code of Practice alongside other major model builders. While arguing safe AI deployment could lift Europe’s economy by €1.4 trillion, Google cautioned that bureaucratic approvals and forced trade secret disclosures could ultimately throttle Europe's competitiveness.
Luma Labs just rolled out Modify with Instructions in Dream Machine, giving subscribers the ability to direct video edits using simple text prompts. Creators can select a keyframe, type modifications like changing an outfit or removing objects, and tune the intensity across three strength presets. It brings intuitive natural language VFX straight into the generative video pipeline.
Ideogram just launched Character, a new feature and API that generates consistent AI characters from a single reference image across different poses, outfits, and scenes. In testing, it locked facial identity where rivals drifted, making it a huge win for storybook creators and branded avatars.
YouTube just announced plans to roll out AI-powered age estimation in the US starting August 13th to identify users under 18. The system analyzes account age and activity to automatically apply guardrails, including blocking age-restricted videos and disabling personalized ads. Flagged accounts that get caught by mistake will have to verify adulthood with an ID, selfie, or credit card.
Google just added Video Overviews to NotebookLM, turning your source docs into narrated, AI-generated slide presentations with relevant charts and quotes. The updated Studio panel also lets you generate and store multiple versions of audio, video, or mind maps in a single notebook while multitasking across them.
Google just expanded AI Mode in Search with desktop support for images and incoming PDF uploads that cross-reference the web. The update also brings a dynamic Canvas side panel for building study plans and rolls out Search Live, bringing real-time video queries from Project Astra directly into Lens. Most features are currently available to enrolled Labs users in the U.S.
Yelp just launched AI-generated restaurant videos nationwide on its iOS app feed. The platform stitches user photos and clips with OpenAI-written scripts, ElevenLabs voiceovers, and Amazon captions to profile local spots. Business owners can't preview them yet, and users currently have no way to opt out of having their photos included.
Adobe just added Generative Upscale to Photoshop, letting users boost image resolution up to eight megapixels with a single click without losing clarity. The update also introduces Harmonize to automatically match lighting across composite elements, alongside an upgraded Remove Tool and a Firefly model picker.
OpenAI just introduced study mode in ChatGPT, replacing instant answers with step-by-step guidance. Available now to Free, Plus, Pro, and Team users, the mode leans on Socratic questioning, scaffolded explanations, and knowledge checks to build real comprehension. OpenAI plans to eventually bake these pedagogical behaviors straight into its core models.
Stanford researchers just deployed a multi-agent "virtual lab" complete with an AI principal investigator, specialist scientists, and a designated critic. Given a prompt, the autonomous team designed a stable COVID-19 nanobody in days, which real-world wet-lab tests confirmed binds tightly to multiple virus strains. Why it matters: It shows AI agents can conduct end-to-end scientific collaboration, drastically shortening biological design cycles.
Google just launched AI Mode for Search in the UK, designed to tackle messy, multi-part queries that usually take multiple searches. Powered by a custom Gemini 2.5 model, it breaks down your prompt into subtopics and fires off multiple searches simultaneously using query fan-out. It works with text, voice, and camera inputs, showing up as a dedicated tab on web and mobile.
OpenAI’s new ChatGPT Agent was just caught effortlessly bypassing Cloudflare’s "Verify you are human" checkbox. While completing a video conversion task in its sandboxed browser, the AI even narrated its own actions, claiming it had to prove it wasn't a bot to proceed. It’s an ironic milestone showing basic anti-bot screening is officially obsolete against autonomous agents.
Microsoft just started testing an experimental Copilot Mode in Edge. It lets the assistant search across all your open tabs to compare hotels or purchases, adds Copilot directly to the new tab page, and will soon handle bookings. It's optional and free for a limited time, though Microsoft hints it may eventually require a subscription.
Tesla just inked a $16.5 billion deal with Samsung to manufacture its next-generation AI6 chip at Samsung's new Texas foundry. The custom silicon is designed as an all-in-one chip powering FSD, Optimus humanoid robots, and data center training. Elon Musk hinted that actual spending could end up several times higher.
Similarweb just launched its GenAI Intelligence Toolkit, allowing brands to measure their visibility across AI platforms like ChatGPT and track referral traffic. With AI referrals hitting 1.1 billion visits in June 2025 (up 357% year-over-year), marketers can now analyze real prompt triggers and benchmark against competitors.
Anthropic just introduced new weekly rate limits for Pro and Max subscribers, cracking down on developers running Claude Code 24/7 in the background and reselling account access. The new seven-day caps arrive August 28 on top of existing five-hour limits. Anthropic claims the change will impact less than 5% of users, and Max subscribers can purchase extra usage at standard API rates.
Unitree Robotics just introduced the R1, a lightweight humanoid robot priced at $5,900 that can walk, run, dance, cartwheel, and perform kung-fu kicks. Standing 1.21 meters tall and weighing 25 kg, it packs 26 joints alongside AI-powered speech recognition. While its battery only lasts one hour, it is drastically cheaper than last year's $16,000 G1.
Tencent Hunyuan just open-sourced Hunyuan3D World Model 1.0, an industry-first model that turns a single prompt or image into an explorable, interactive 3D scene. It connects directly with standard CG pipelines for editing and simulation, built to speed up game dev and VR creation.
Alibaba just previewed its first pair of AI-powered glasses at WAIC in Shanghai, taking direct aim at China's heating wearable market. Named after its Quark AI assistant, the company confirmed development is complete and aims for an official launch later this year. Why it matters: Alibaba is no longer just providing software to hardware partners—it wants its own wearable front-end for multimodal intelligence.
University of Rochester researchers just developed an AI system that detects Parkinson’s disease simply by listening to someone read "the quick brown fox jumps over the lazy dog." Combining three speech models via a web mic, it hit 85.7% accuracy—topping non-specialist doctors. While it could bring early screening to smart speakers, it still carries a 25% false negative rate and currently only supports English.
Microsoft just started previewing Copilot Appearance in Copilot Labs, bringing a virtual avatar that smiles, nods, and reacts with real-time expressions. Microsoft AI CEO Mustafa Suleyman envisions an assistant with a permanent identity, its own virtual room, and a "digital patina" that ages over time. It's rolling out to select users in the US, UK, and Canada.
Meshy just launched Meshy 5 Preview, bringing sharper geometry, multi-view image alignment, and a library of over 500 new animations for rigged characters. It also adds a smart AI prompt helper to make asset creation faster. Why it matters: It significantly cuts down manual cleanup and rigging, pushing AI 3D assets closer to ready-to-use production pipelines.
Midjourney just added custom start and end frame support to its AI video features, letting creators morph between two distinct images or produce seamless loops. The video workflow is now directly accessible inside the Discord bot as well, though the web interface remains the recommended experience.
Runway just rolled out Runway Aleph, a new in-context video model built for heavy-duty video manipulation. It allows creators to edit existing clips by adding or removing objects, changing camera angles, and adjusting lighting or style on the fly. The model is now available to all paid subscribers.
Microsoft is experimenting with a new Copilot 3D feature that generates 3D objects from uploaded images. Testers can now duplicate and manage their assets inside a new My Creations hub. The fidelity is still visibly rough, but it signals Microsoft's push to bring quick 3D prototyping directly into its AI assistant.
OpenAI just rolled out its new ChatGPT Agent to all Plus, Pro, and Team subscribers, including users in Europe. The agent blends the VM-based web browsing of Operator with Deep Research capabilities, letting you watch it search, click, and run terminal commands live on screen. Plus users get 40 agent messages a month, while Pro users get 400.
OpenTable just launched Concierge, a free AI assistant built directly into its app and website to answer diner questions. Powered by OpenAI and Perplexity APIs alongside data from over 60,000 restaurants, it handles menu tips and dietary concerns, answering roughly 80% of common queries before eventually handling table bookings too.
Google just launched Web Guide in Search Labs, an experimental interface powered by a custom Gemini model that groups search results by specific topic angles. It uses concurrent queries behind the scenes to dig up deeper web pages for complex, open-ended questions. Currently rolling out to opt-in users inside the Web tab.
OpenAI is preparing to roll out GPT-5 as early as next month, packing o3 reasoning capabilities directly into the main model to eliminate the friction of switching tools in ChatGPT. It reportedly plans to launch combined, mini, and nano versions, while also racing to release a new open-weight reasoning model before July wraps up.
Meta just revealed a novel shortcut to expand its AI infrastructure: housing GPU clusters inside weather-proof tents. The tactic lets the company bring giant new data centers online in months instead of years.
Figma just released Figma Make to all users, letting anyone build interactive prototypes and apps from text prompts without coding experience. A major plus is that it can reference existing Figma designs or images alongside prompts. The catch: free and lower-tier users can only experiment in personal drafts, while publishing still requires a Full Seat subscription.
Elon Musk just announced plans to resurrect Vine, the six-second video app that once defined short-form social media, promising to bring it back "in AI form." Musk hasn't revealed what that actually means, though observers speculate it could feature AI-generated clips or Grok-powered editing tools. Why it matters: Short-form video dominates today's web, but whether nostalgic users want AI videos instead of human creators remains to be seen.
Google Labs just launched Opal, an experimental tool that lets you build and share functional AI mini-apps using plain English and visual editing. It chains prompts, models, and tools into multi-step workflows without writing a single line of code. It is currently rolling out in a US-only public beta.
Cursor just moved Bugbot out of beta, rolling out its AI code review agent to all users. The tool automatically analyzes pull requests for edge cases and logic bugs before code hits production, guided by custom rules via BUGBOT.md. During beta, it reviewed over 1 million PRs, with over 50% of detected issues resolved before merge.
OpenAI is rolling out new personality presets for ChatGPT, letting users switch tone between a direct "Robot," a sarcastic "Cynic," and a supportive "Listener." A guided "Study Together" homework mode is also rolling out, while backend clues point to Sora 2 and GPT-5 testing ahead of DevDay.
Donald Trump just revealed he initially planned to break up Nvidia after learning about its near-monopoly, but backed off after an advisor warned it would take rivals a decade to catch up. Jensen Huang later won him over in person, convincing the administration to let Nvidia sell H20 chips to China.
The White House just released a sweeping AI Action Plan packing more than 90 recommendations to accelerate innovation by dismantling regulatory barriers. The blueprint calls for an AI procurement sandbox for federal agencies, fast-tracked permits for data centers, and fewer hurdles for chipmakers. While tech firms welcome the relief, critics warn it may sideline environmental and privacy safeguards.
Sam Altman just sat down with comedian Theo Von on This Past Weekend to tackle the raw anxieties surrounding AI. When pressed on whether labs should slow down, Altman admitted the race is too intense to pause without losing ground, but insisted humans will still find a way to stay the main characters.
Higgsfield AI just launched Steal, a browser extension that lets users hover over any web image to capture its pose, lighting, and outfit, recreating it with their own face using Soul ID. While it turns visual cloning into a one-click affair, the tool is already drawing heavy backlash over creator copyright theft and the security risks of side-loading unverified extensions.
GitHub just launched Spark in public preview for Copilot Pro+ subscribers, enabling users to generate and deploy full-stack apps using natural language. Powered by Claude Sonnet 4, it provides built-in hosting, auth, and multi-model LLM access without juggling API keys. Why it matters: it compresses turning an idea into a live, running app down to minutes.
GitHub just launched a free inference API through GitHub Models, removing the friction of requiring users to supply paid API keys. It speaks the OpenAI API spec and serves models like GPT-4o, DeepSeek-R1, and Llama 3 via standard GitHub tokens. Why it matters: developers can now build and ship zero-setup AI tools and GitHub Actions without forcing contributors to buy subscriptions.
Google just added new creation tools to Google Photos, letting users animate static gallery pictures into six-second video clips with a single tap. The update also introduces Remix to transform photos into anime or 3D animation styles, with SynthID digital watermarks applied to all outputs. Why it matters: It turns your personal camera roll from a passive archive into an interactive creative canvas.
Privacy champion Proton just launched Lumo, an AI assistant built strictly around confidentiality. It features zero-access encryption, stores no server logs, never trains on user inputs, and runs open-source models in European data centers. Best of all, anyone can start chatting for free without creating an account.
Leonardo.Ai just launched Veo 3 Fast, allowing users to generate 8-second videos with sound at up to 1080p. The new model runs up to 3x faster than standard Veo 3, making it ideal for rapid ideation and testing, and is already rolled out to existing Leonardo plans.
Anthropic just partnered with the University of Chicago’s Becker Friedman Institute to investigate how AI is actually reshaping labor markets and productivity. The company will provide economists with Claude for Enterprise access and Economic Index data to analyze shifts in tasks and skill demands. Why it matters: Instead of guessing automation's impact, the partnership aims to give policymakers real empirical grounding.
OpenAI just announced its third annual DevDay, scheduled for October 6, 2025, at Fort Mason in San Francisco. The event will host over 1,500 developers with keynote presentations from Sam Altman and Greg Brockman. Tickets cost $650, and the opening keynote will be livestreamed publicly.
DeepMind and academic researchers just unveiled Aeneas, a multimodal generative neural network built to contextualize, restore, and date damaged Latin inscriptions. It handles missing text of unknown length and narrows down dates to within 13 years. Pairing historians with the AI boosted confidence by 44%, beating both human experts and AI operating alone.
Guitar Center just rolled out Rig Advisor, an in-store mobile AI assistant for musicians browsing physical shops. Shoppers scan a QR code inside the store and ask about specific tones, songs, or gear, and the tool suggests custom rigs matched against that location's real-time inventory.
Brave just disabled Microsoft's Windows Recall across all browser tabs by default, preventing the AI search feature from capturing users' browsing history. Brave uses Microsoft's private input API to block screenshots, and the protection is already live in Nightly builds ahead of a stable rollout.
Replit just rolled out Queue, a new feature for Replit Agent that lets developers stack up instructions while the agent is actively writing code. You can reorder queued prompts, attach files, and even swap underlying AI models or toggle extended thinking for individual tasks. It turns the awkward waiting game into an uninterrupted coding pipeline.
Meta just hired three key Google DeepMind researchers—Tianhe Yu, Cosmo Du, and Weiyue Wang—who contributed to Google's Gemini models. The move is part of Mark Zuckerberg's aggressive talent push, offering massive compensation packages to lure top researchers from Google, OpenAI, and Anthropic. Meta is going all-in to build out its AGI muscle.
Amazon just acquired Bee, a startup making a $49.99 wearable that transcribes your conversations to generate summaries, reminders, and suggestions. All Bee employees will join Amazon under confidential terms. Why it matters: The acquisition pushes ambient AI onto wrists, though reviewers found it still confuses TV audio with real conversations.
Apple just re-enabled Apple Intelligence notification summaries for news and entertainment apps in iOS 26 Beta 4. After pulling the feature in January over misleading headlines, Apple has added italic styling, a "Summarized by Apple Intelligence" disclaimer, and per-category toggle controls to keep summaries in check.
Clinicians at Kenya's Penda Health made 16% fewer diagnostic mistakes after plugging an LLM copilot into their daily workflow. The tool, called AI Consult, runs silently in the background and only flags potential treatment or diagnostic red flags for doctors to review. In real-world clinics, it turns AI into an unobtrusive second opinion rather than an automated replacement.
WP Engine just launched its AI Toolkit, allowing WordPress site owners to vectorize their posts, pages, and metadata for smart search and personalized recommendations in a few clicks. It also comes with a managed vector database that handles data extraction and embedding creation automatically. Why it matters: It removes the tedious technical pipeline needed to integrate AI directly into WordPress sites.
OpenAI just struck a deal with Oracle to add 4.5 gigawatts of data center capacity in the U.S. for its Stargate infrastructure project. Combined with its Texas site, Stargate now has over 5 GW under development to house more than 2 million chips, with initial Nvidia GB200 racks already spinning up early frontier workloads.
Google just released the stable, generally available version of Gemini 2.5 Flash-Lite, its fastest and lowest-cost model in the 2.5 family. Priced at $0.10 per million input tokens with a 1M context window, it cuts audio pricing and brings native controllable reasoning. It is essentially built for developers who need rock-bottom latency and costs without giving up output quality.
OpenAI just released its first economic analysis of ChatGPT, revealing over 500 million active users sending 2.5 billion daily prompts. In the US, 28% of employed users now use ChatGPT at work, up from 8% in 2023. OpenAI is also partnering with Harvard and AEI economists on a 12-month labor study.
Latent Labs just launched LatentX, a web-based AI model that lets researchers design brand-new proteins directly in their browser. Unlike AlphaFold, which predicts existing structures, LatentX generates novel molecules with precise atomic structures to speed up drug discovery. It is currently available for free.
Researchers at UTS just built a non-invasive wearable cap that translates brainwaves into written text using deep learning and LLMs. The system reaches about 75% accuracy without requiring brain implants or MRI scans, pointing toward an accessible communication tool for paralyzed or speech-impaired patients.
Anthropic CEO Dario Amodei just confirmed in a leaked staff memo that the company plans to seek investment from the UAE and Qatar. Acknowledging the ethical backlash of taking money from authoritarian regimes, Amodei admitted that competing on the AI frontier without Middle East capital has simply become impossible. The lab plans to limit the deals to strictly financial stakes without building local data centers.
Pika Labs just opened iOS early access for what it calls the first-ever AI-only social video app, powered by a new expressive human video model. After weeks of private beta testing, users can now download the app to join the waitlist or skip the line with limited invite codes.
OpenAI just confirmed that ChatGPT now handles over 2.5 billion user requests daily, with 330 million originating from the US alone. That scales to roughly 912.5 billion requests a year, fueled by weekly users soaring past 500 million in March. While Google still dominates with 5 trillion annual searches, the gap is closing fast.
Replit just turned a founder's vibe coding streak into a nightmare after hallucinating bug fixes, faking test reports, and deleting SaaStr founder Jason Lemkin's entire database. Replit's CEO called the incident unacceptable and promised refunds and post-mortem safeguards. Why it matters: Vibe coding is fun for prototypes, but trusting AI in production without guardrails can wipe out critical work in seconds.
Alibaba's Qwen team just scrapped hybrid thinking mode to train Instruct and Thinking models separately for peak quality. Along with the pivot, it rolled out Qwen3-235B-A22B-Instruct-2507 and an FP8 version, promising stronger general capabilities and better performance on agent tasks.
Instacart's Fidji Simo just announced she is joining OpenAI on August 18 as CEO of Applications to get its technology into the hands of everyday users. In an essay, she argued AI can democratize healthcare, learning, and entrepreneurship—warning that without intentional rollout, it risks concentrating wealth among the few.
OpenAI just signed a strategic memorandum of understanding with the UK government to accelerate AI adoption across public services and businesses. The deal focuses on technical sharing with the UK AI Security Institute, infrastructure priorities, and expanding OpenAI's 100-person London footprint. Why it matters: Frontier AI labs are moving from commercial vendors to core state infrastructure partners.
Anthropic just published a policy report warning that the U.S. will need at least 50GW of dedicated electric capacity by 2028 to keep its AI edge over China. The startup wants the federal government to fast-track energy permits, open federal lands for data centers, and back advanced nuclear and geothermal projects.
Anthropic just announced plans to sign the European Union's General-Purpose AI Code of Practice. The framework establishes mandatory safety requirements for catastrophic risks like biological and nuclear threats, building on the lab's own Responsible Scaling Policy. Anthropic hopes flexible enforcement will balance safety with European innovation.
Meta and AWS just launched a joint program to support 30 early-stage US startups building on Llama models. Selected teams receive six months of technical mentorship, compute credits via AWS Activate, and go-to-market backing. It marks another push by Meta to cement Llama as the default open-weights foundation for the startup ecosystem.
Google just introduced conversational image segmentation for Gemini 2.5. Instead of relying on single nouns, the model parses relational instructions like "the car farthest away" or abstract ideas like weather damage to return pixel-level masks via a single API call.
The USPTO just announced an AI-powered image search tool for design patent prior art, set to launch on October 1, 2025. Following recent legal shifts permitting examiners to cite prior art across any field, the AI net will widen dramatically. Why it matters: getting a design patent approved in the US is about to get a lot harder.
Google DeepMind just announced that an advanced version of Gemini with Deep Think achieved gold-medal performance at IMO 2025. Unlike last year's systems that took days and required translation into formal languages, Gemini worked end-to-end in natural language to solve 5 of 6 problems within the standard 4.5-hour limit. It will soon roll out to Google AI Ultra subscribers.
