Genie 3: A new frontier for world models ↗
DeepMind unveils Genie 3, a groundbreaking AI that generates fully interactive, real-time 720p worlds at 24 FPS. From volcanic landscapes to enchanted forests, users can navigate and prompt dynamic changes—like weather or new objects—for minutes-long simulations. Genie 3 paves the way for advanced agent training and creative applications, while prioritizing responsible development.
Only 2 days left to save $675 on your TechCrunch Disrupt 2025 ticket ↗
TechCrunch Disrupt 2025 celebrates its 20th anniversary from October 27–29 at Moscone West in San Francisco. Join 10,000+ founders, investors and tech pros across AI, Space, Public Markets and Startup Battlefield stages. Network via interactive sessions, pitch live or scout the next big idea. Lock in savings of up to $675 by August 6 at 11:59 p.m. PT.
US adds OpenAI, Google and Anthropic to list of approved AI vendors for federal agencies ↗
The U.S. General Services Administration (GSA) has added Google, OpenAI, and Anthropic to its Multiple Awards Schedule, a streamlined contracting platform for civilian federal agencies. These vendors, vetted on security and performance, can now offer AI services under pre-negotiated terms. The move follows executive orders boosting data center energy supplies and mandating bias-free AI use.
Should big tech be allowed to mine Australians’ text and data to train AI? The Productivity Commission is considering it ↗
Australia’s Productivity Commission has proposed reviewing copyright and privacy laws to let tech firms mine text and data for AI training. Its interim report recommends exemptions for text and data mining—or a new compensation scheme for content creators—to boost innovation while protecting rights. Commissioners warn that poorly scoped rules could risk misuse by “bad actors,” but they caution against sweeping AI-specific legislation, urging regulators to adapt existing frameworks. They estimate AI could add A$116bn to the economy in a decade, lifting average real wages by around A$4,300.
AI-powered fintech Alaan raises $48M, one of the largest Series A rounds in MENA ↗
Alaan, an AI-powered spend management platform founded by ex-McKinsey consultants in Dubai, raised $48M in a Series A led by Peak XV, marking one of the largest fintech funding rounds in MENA. The platform integrates Apple Pay and OpenAI to automate expense reconciliation and VAT extraction, saving finance teams over 1.5M manual hours. Profitable and processing 2.5M+ transactions for 1,500+ teams, Alaan is scaling into Saudi Arabia, using the new funds to boost sales, customer success, and AI-driven automation.
Google says its AI-based bug hunter found 20 security vulnerabilities ↗
Google’s LLM-powered bug hunter Big Sleep, built by DeepMind and Project Zero, spotted 20 security flaws in popular open-source tools like FFmpeg and ImageMagick. While human experts verify each finding, Big Sleep autonomously detected and reproduced the bugs, signaling a leap in automated vulnerability hunting. Other AI tools like RunSybil and XBOW are also emerging, though some reports can turn out to be false alarms. Google is withholding details until fixes are ready.
OpenAI says ChatGPT is on track to reach 700M weekly users ↗
ChatGPT’s user base is set to hit 700 million weekly active users, up from 500 million in March and a four-fold increase since last year. The surge follows an upgraded GPT-4 image generator—over 130 million users created 700 million images—and has driven business subscriptions from 3 million to 5 million. Users now spend an average of 16 minutes daily on the app.
Grok Imagine, xAI’s new AI image and video generator, lets you make NSFW content ↗
Elon Musk’s xAI has launched Grok Imagine on iOS for SuperGrok and Premium+ X subscribers, offering 15-second AI videos with a “spicy mode” that allows partial nudity and NSFW imagery. While extreme prompts are blurred or moderated, users can generate semi-nude content and stylized animations. Grok Imagine competes with Google DeepMind, OpenAI and others, though its human visuals still sit in the uncanny valley. Musk promises daily improvements amid concerns over misuse and unintended consequences.
OpenMind wants to be the Android operating system of humanoid robots ↗
OpenMind is building OM1, an open, hardware-agnostic operating system for humanoid robots, akin to Android. Founded by Stanford professor Jan Liphardt, the company unveiled FABRIC, a protocol enabling robots to verify identities and share context, speeding collective learning—like instant language acquisition. Starting September, OpenMind plans to ship 10 OM1-powered robotic dogs for home testing, iterating quickly based on user feedback. After raising $20 M led by Pantera Capital, the startup aims to refine its OS through real-world trials, forging seamless human-machine collaboration.
Demis Hassabis on our AI future: ‘It’ll be 10 times bigger than the Industrial Revolution – and maybe 10 times faster’ ↗
DeepMind chief Demis Hassabis predicts AI will dwarf the Industrial Revolution in scale and speed, promising radical abundance—from medical breakthroughs thanks to AlphaFold to advances in energy and materials. A former chess prodigy and self-taught coder, he co-founded DeepMind, sold to Google in 2014, and now foresees artificial general intelligence within five to ten years. While optimistic about productivity and prosperity, he acknowledges challenges like energy demands, job displacement and equitable distribution.