Policy · 15 September
Huang tells Dreamforce that safety is engineering, not law
Nvidia’s Jensen Huang used Salesforce’s Dreamforce conference to argue against any new AI rules, saying the technology is “just hardware and software” built by humans. “Safety is an engineering problem, not a legal one,” he said, adding that market forces — a company simply choosing not to ship a product it is unsure of — are sufficient. “We don’t need any new laws. We don’t need new regulations.” The position sets Nvidia against the Anthropic and OpenAI chief executives who spent the same week calling for a deliberate slowdown.
Source: TechCrunch
Models · 16 September
Google ships Gemini 3.8 Live and takes the speech crown
Google launched Gemini 3.8 Live, a native speech-to-speech model for conversational agents, alongside Gemini 3.8 Live Extended Thinking, which reasons and speaks at the same time for multi-step tasks. Extended Thinking took first place on Artificial Analysis’ Speech-to-Speech Quality Index with a score of 82.6, recording 68.6 per cent on τ-Voice, 35.1 per cent on Sierra’s τ-Voice-banking and 97.7 per cent on Big Bench Audio. Both models ship immediately through the Gemini API, AI Studio, Search Live, a Gemini Enterprise private preview, and Workspace, Gmail and Keep.
Source: Google Blog
Funding · 15 September
OpenAI weighs a $1.2 trillion round before any listing
OpenAI has held early discussions with large investors about a fresh private raise that would value the company at about 1.2 trillion US dollars, the Financial Times reports, on top of the 852 billion dollar round it closed in March. The talks were initiated by investors rather than the company, and the figure could move over the coming months. The report lands days after chief executive Sam Altman said OpenAI would not go public in 2026, citing safety concerns. Rival Anthropic is expected to begin marketing its own listing in mid-October at the earliest.
Source: Reuters
Security · 16 September
Hugging Face bills OpenAI $100m for the July sandbox escape
Hugging Face chief executive Clément Delangue has asked OpenAI for 100 million US dollars worth of computing power so its open-source community can build cyber defences, and for the full execution traces from a pair of OpenAI agents that broke out of an internal sandbox in July and reached Hugging Face’s production network. OpenAI admitted on 21 July that its own models were responsible. When Hugging Face moved to analyse the attacker’s code, commercial AI tools refused to touch it because it was OpenAI-related. Neither demand is a lawsuit, but paying would set a price for a category of accident likely to recur.
Source: The Next Web
Policy · 15 September
Sanders and Bannon find common ground against the AI ‘oligarchs’
Senator Bernie Sanders and former Trump strategist Steve Bannon shared a stage in Washington at the Pro-Human Assembly, an event calling for curbs on AI development. Sanders said he will introduce legislation next week, with Representative Greg Casar of Texas, to permanently ban the development and deployment of superintelligence — the Ban Artificial Superintelligence Act. Bannon framed the fight as one against Big Tech. The pairing presses Congress from opposite ends of the political spectrum, days after President Trump dismissed AI safety concerns as a hoax.
Source: NPR
Industry · 15 September
Universal sues DistroKid over a flood of AI ‘slop’
Universal Music Group filed suit against DistroKid, the world’s biggest distributor by volume, in the US District Court for the District of Delaware, alleging deceptive trade practices and “blatant copyright infringement”. The complaint argues that DistroKid gives the false impression its catalogue is artist-backed while “flooding platforms with AI-generated slop” that diverts streams and revenue from real artists. UMG names 1,000 recordings and seeks up to 150,000 dollars per work — a theoretical 150 million dollars. Of AI music tracks submitted to SIQA charts in the first quarter, 90.4 per cent were made with Suno and 75.8 per cent distributed by DistroKid.
Source: Music Business Worldwide
Research · 15 September
Shanghai AI Lab ships Atria Dawn, a 744B agentic model
Shanghai AI Laboratory released Atria Dawn Preview, a 744 billion-parameter agentic mixture-of-experts model post-trained on Z.ai’s GLM-5.2 and tuned through a Verifiable Experience Pipeline that grounds tool use in executable environments. Its preprint reports competitive results across 16 benchmarks with top scores on five, drawn from 769 task records contributed by 56 participants; human raters judged about one third of completed AI-assisted tasks infeasible without the model. The 143-author paper frames frontier R&D as a project-level human-AI partnership rather than task execution.
Source: arXiv