The CFTC exempted “passive software” providers from broker registration as long as they never hold user assets, per Bloomberg’s Lydia Beyoud, a move that opens the door wider for crypto and prediction-market trading platforms. World rolled out World Money, a self-custodial financial “super app” live in more than 150 countries that bundles stablecoin payments, trading, earning, and virtual accounts, The Block’s Jason Shubnell reported. Reuters sources say Apple plans to bring Apple Pay to India next month through Axis Bank’s credit cards, tapping the country’s fourth-largest card issuer. Meta pushed its Muse AI agent onto Mac after earlier iOS, Android, and web releases, giving it the ability to manage files and pull from other apps, The Verge’s Jay Peters noted. And in a story that will make security teams wince, the Wall Street Journal’s Robert McMillan reported that researchers in an OpenAI bug bounty program breached the company’s GitHub monorepo using a cybersecurity-tuned version of Opus 4.8 and Opus 5.
On the model front, PrismML released Bonsai 2 27B, a compression of Alibaba’s Qwen3.8 27B down to 5.9 GB — small enough to run on a phone — while holding onto 98.2% of the original’s benchmark performance, TechCrunch’s Julie Bort wrote. A federal judge denied OpenAI’s request to see SpaceXAI’s confidential antitrust settlement with Apple, ruling the materials irrelevant after review, according to Politico. Dwarkesh Patel’s podcast featured OpenAI researcher Noam Brown on multi-agent systems, AI’s progress on the Navier-Stokes problem, and the gap between internal and external model capability. The UN said it’s partnering with Google on a System Data Commons that lets users query statistics across UN agencies in natural language, TechCrunch’s Jagmeet Singh reported.
Anthropic had a busy stretch: Bloomberg’s Shirin Ghaffary reported the company saying Claude now “leads” 26% of its internal AI R&D work, up from just 1% in March, and collaborates on more than 90% of it under close human direction. Claude itself announced a redesign of Claude projects, letting people describe work once and have it managed across parallel threads, starting in Claude Code. Separately, Anthropic laid out proposed metrics for tracking frontier AI development — how much R&D work AI itself is doing, how well agents are being overseen, and how compute gets allocated.
OpenAI launched Astra for Law, pairing GPT-6 Astra with a legal search index and analysis tooling, rolling out initially to select law firms. On the more skeptical side of the AI debate, Andrew Ng told Bloomberg’s Micah Barkley that renewed AI extinction warnings are “much more science fiction than science” and that the latest wave of alarm is likely PR aimed at shaping regulation. A New York Times court filing referenced in the Financial Times quoted ChatGPT’s product head describing publishers as facing an “existential threat” from AI, while a Microsoft director reportedly called AI training “an astonishing theft.” Pew Research surveyed 42,151 adults across 36 countries plus 8,607 more in the US and found people in 34 of 37 countries expect AI to cause net job losses rather than net job growth — a sentiment echoed in Bloomberg’s Francesca Maglione’s analysis showing Bay Area tech employers filed layoff notices for more than 14,500 workers in the year to June, with software engineer demand down 42% since 2022.
On the money and infrastructure side, Bloomberg’s Carmen Arroyo reported SpaceX has explored buying customer and operational data from distressed or defunct startups as a cheaper source of AI training data. Dutch chipmaker Nexperia struck a manufacturing and packaging deal with Tata in India, deepening its split from Chinese parent Wingtech, Reuters’ Toby Sterling wrote. Bloomberg sources say the White House’s cyber director is drafting an executive order to stand up a government-led cybersecurity research incubator. The UK’s FCA raided three illegal peer-to-peer crypto operations in coordination with tax authorities and police, CoinDesk’s Olivier Acuna reported, signaling the end of its lighter-touch crypto stance. The Information’s Stephanie Palazzolo reported OpenAI staff expect the Hodge Conjecture — a Millennium Prize Problem — to fall relatively soon, following the firm’s work on Navier-Stokes. King Charles hosted tech leaders including Jensen Huang, Demis Hassabis, and OpenAI CFO Sarah Friar in Scotland to discuss AI risk, with Huang calling for safety testing. Lucid and Bolt signed an initial agreement to deploy more than 25,000 robotaxis across Europe on Lucid’s EV platform, targeting a 2028 launch, per the Financial Times. Comp AI, which uses agents to draft security policies and run continuous compliance monitoring, raised a $34 million Series A, TechCrunch’s Dominic-Madori Davis reported. The SEC unveiled a five-year Innovation Exemption for tokenized stock trading platforms, freeing them from many rules that apply to traditional exchanges, Reuters’ Hannah Lang wrote.
A former US State Department envoy told the New York Times that US-China AI safety talks won’t produce a breakthrough treaty, arguing Beijing is prioritizing strategic advantage. 9to5Mac’s Chance Miller reviewed the Apple Watch Series 12 and Ultra 4, praising new colors and materials and calling the Health Sensing System a big upgrade, though many features are still pending. DoorDash extended same-day delivery to every US Costco location for members, a day after Uber expanded its own Costco delivery service from 17 to 47 states. Huawei’s chairman, Eric Xu, said Chinese AI researchers need to accelerate development to properly “see the dangers” of AI, a pointed contrast to US calls for a slowdown, the Financial Times’ Eleanor Olcott reported. The New York Times’ David Streitfeld traced how Dario Amodei’s essays chart the industry’s shifting messaging on AI risk, from wariness of regulation to calls for third-party evaluation. The EU detailed its Kids Act, which would mandate age verification to keep under-15s off social platforms without parental consent and ban under-13s outright, per the Wall Street Journal’s Edith Hancock.
Pew also found that, for the first time, Democrats are more worried than Republicans about AI’s growing role in daily life — 56% to 49% — with 75% of Democrats specifically worried about job losses. At India’s Semicon event, Narendra Modi courted investors, with Bloomberg’s Sankalp Phartiyal reporting pledges of billions in investment from Applied Materials and Lam Research. The Wall Street Journal profiled Jacob Coxon, the British researcher who left Anthropic after a string of AI security incidents, saying he was surprised by the intensity of the response. Ars Technica’s Andrew Cunningham reviewed macOS 27 Golden Gate, finding it fixes many Tahoe-era Liquid Glass issues and adds useful non-AI features, though it drops Intel support and eats a lot of storage. Bloomberg’s Alice Lassman examined how AI beauty tools like Qoves and UMax are commercializing facial-analysis algorithms that assess geometric proportions and balding to recommend treatments. Reuters reported that US and Chinese experts are proposing nuclear-style AI safeguards, including a dedicated military hotline, ahead of planned bilateral AI talks. Snap is pitching its $2,195 Specs as an enterprise product, striking deals with Salesforce, Amazon, and Nvidia for visual-overlay use cases alongside its consumer push, Wired’s Boone Ashworth wrote. Data center developer Crusoe raised $3.9 billion co-led by Atreides, Valor, and Mubadala at roughly a $30.9 billion post-money valuation, betting on factory-built data centers, per the Wall Street Journal’s Anissa Gardizy.
Further back, the Financial Times examined how quickly AI is being woven into warfare, warning that the speed and scale of AI-assisted targeting can raise the risk of errors. The Verge’s Yael Grauer covered research from Anthropic and others detailing a network of roughly 28 dating scam apps that used LLM-generated replies — including from models like Claude — to catfish thousands of victims. PitchBook data cited by the Financial Times showed brain-computer interface startups have raised more than $1 billion in 2026 alone, versus $1.56 billion across the previous four years combined, with Neuralink the best-funded. Huawei said it plans to debut its Ascend 960DT AI chip in the first quarter of 2027 and the 960PR in the third quarter, alongside an architecture that would let up to a million processors act as a single system, Reuters reported. Google, Nvidia, and Emerald AI launched an AI Energy Management Alliance to push data centers that dynamically adjust power draw based on grid conditions, Axios’ Amy Harder wrote. TechCrunch’s Sarah Perez reported documents showing Automattic’s interim CEO Mark Davies and its chief legal officer signed each other’s severance agreements, worth $8.15 million, during Matt Mullenweg’s brief ouster. Bloomberg reported that Sam Altman and Jensen Huang are among the business leaders expected at a White House state dinner for Xi Jinping next week, with Tim Cook also said to be attending. The Financial Times reported that Emulate, a month-old UK world-models startup founded by ex-Google DeepMind staff, is in talks to raise up to $700 million at a $3.7 billion post-money valuation. Bloomberg also reported that Manus is close to closing a $500 million round at a $4 billion valuation, a sign of rebounding confidence after Beijing unwound Meta’s short-lived $2 billion buyout of the company. And the New York Times’ Cade Metz looked at how AI startups like Inherent and Recursive Superintelligence are chasing the tools needed to push AI systems toward recursive self-improvement.