📰 Welcome back!

The sun is shining in the world of AI this week, after capping-off a late summer belter thanks to OpenAI’s launch of Sora 2.

Thanks to one of the best generative video models on the market, the shit-posting has begun in earnest and looks set to pour petrol on a million copyright infringement cases in the near future.

It’s also been a week of high-powered deals that would make Gordon Gecko blush, including those between Microsoft and Anthropic; Anthropic and everyone it stole content off; and OpenAI and South Korea (basically).

Anyway, down to business. Enjoy.

🚀 What we’re covering today…

• 🎥 OpenAI launches Sora 2 – and it’s impressive

• 👩🏼‍⚖️ Anthropic get hit with $1.5B copyright settlement

• 🪟 Microsoft opens Copilot up to Claude

• 🇰🇷 OpenAI strikes $500B deal with Samsung

• 🔎 California introduces AI regulations

• 😩 Trump shares racist AI content AGAIN

• 🤖 Meta will use AI searches to shape ad algos

• 🥷 Tilly Norwood is a psy-op

• 🫂 The “hug my younger self” trend hits in the feels

🔴 Quick Note: We like to cover loads of AI news in our newsletter, so for a better reading experience, we suggest opening this in your browser for the full experience! 

Head to the ‘READ ONLINE’ tab at the top of this email.

👁️ 👁️ What you might have missed

• Finally, OpenAI’s generative AI video model, Sora, has reached a state as god intended: highly proficient at creating highly offensive virtual slop. That’s right, the launch of Sora 2 this week has already been put to good use by the bottom feeders of the internet, showcasing its powerful ability to create bizarro deepfakes, mind-dumbing all-purpose brain-rot, and extremely convincing reproductions of copyrighted materials bound to create an expensive and legally perilous headache for IP owners.
  
  “As a general purpose video-audio generation system, it is capable of creating sophisticated background soundscapes, speech, and sound effects with a high degree of realism,” OpenAI’s launch notes read, and from what we’ve seen, they’re not wrong. And as a successor to the first Sora model, released in 2024, we can now say Google’s Veo3 genAI video model has a legitimate competitor. Needless to say, it shits on Grok’s and Llama’s frustratingly inconsistent image generation capability from great height, too.
  
  Because of the likelihood it will be used for nefarious means, Altman and team emphasized Sora 2’s safeguarding features, which will help curb its use to create deepfakes of real people and the sexualisation of cherished childhood memories, for instance (no one needs to see a Bikini Bottom orgy, ever). Last week’s launch was invite-only, but access to Sora 2 can be imminently accessed via the Sora iOS app, or through ChatGPT. Usage is currently free, with generous limits for now, but access will invariably be limited to subscription plans like other models – give it a go, and share your worst with us.

• More movement on one of the most understated copyright infringement cases of the century: A federal judge has preliminarily approved a $1.5 billion settlement between AI company Anthropic and authors who claim nearly 465,000 books were improperly used to train its AI models. Under the deal, authors and publishers would receive about $3,000 per book included. The settlement covers past works only (not future ones) and mandates a claims process to ensure eligible parties are notified. A slap on the wrist for Anthropic really, a company with an estimated $183B market value after closing a recent round of funding with $13B in September. The ruling does set a significant precedent, however, especially with Apple is staring down the barrel of its own class action lawsuit for allegedly training its OpenELM models on 196,000 pieces of copyrighted works and materials. Ouch.
  

• The Brain Trust at Microsoft has finally opened its arms to the agentic power of its competitors and introduced its Copilot AI assistant to Anthropic’s Claude Opus 4.1 and Claude Sonnet 4 models, alongside OpenAI’s. Users can now select models for different tasks – Opus for complex reasoning and coding, Sonnet for scalable data and content work – within Copilot Studio and the Researcher agent, managed via the Microsoft 365 admin center. Despite unchanged pricing, analysts see this move as key to boosting adoption, which currently lags at just 4% enterprise-wide, and reducing reliance on OpenAI.
  

• Half a trilly, what the helly? OpenAI has struck partnerships with South Korean tech giants Samsung and SK Hynix to supply advanced chips for its ambitious Stargate AI infrastructure project, estimated at $500B. The deals also include collaborations with SK Telecom for “Stargate Korea” data centers, and work on floating data centers to mitigate land and cooling constraints. The agreements emerged in conjunction with OpenAI CEO Sam Altman’s meeting with Korean leaders, emphasizing Korea’s strategic role in AI semiconductor development. It follows the cosmic level of investment OpenAI had agreed with Oracle for cloud services for Stargate, earlier in the year.
  

• California has passed SB 53, also known as the “Transparency in Frontier Artificial Intelligence Act,” requiring large AI companies to publicly disclose safety and security practices, publish updates within 30 days, and adopt whistleblower protections. It also creates a mechanism for reporting “critical safety incidents” to the state’s Office of Emergency Services, and enables the attorney general to enforce compliance. Kind of at odds with Florida’s Senator Ted Cruz’s proposed SANDBOX legislation, which would override the kind of regulation California’s lawmakers are enacting, but this current administration is split down the middle about AI regulation and enforcement. Which leads nicely onto our next story…
  

• Shouldn’t laugh, but the lad just can’t himself, can he? Just hours before a US government shutdown, Trump posted a racially charged, AI-altered video mocking House Minority Leader Hakeem Jeffries and Senate Minority Leader Chuck Schumer. In the clip, Jeffries is shown wearing a sombrero and moustache while Mariachi music plays; Schumer is depicted claiming Democrats support free healthcare for undocumented immigrants. Jeffries condemned the video as “racist and fake” and called it a deliberate distraction. Think we could actually publish an entire newsletter to Trump’s greatest AI hits…
  

  

  😩 Source: MSNCB

  
  

• Meta will begin using users’ interactions with its AI tools – via text or voice – to tailor content and ads across Facebook and Instagram starting December 16, 2025. Meta says it won’t use AI chat data on sensitive topics like religion, health, politics or race for targeting. Users won’t have an opt-out, but the rollout excludes the UK, EU, and South Korea initially. Conversations about topics such as gooning, for example, could influence which posts or ads you see for the Flashlight Turbo 5000. But you see them already don’t you, you dirty little fuck-pig?
  

• Elon Musk’s feuds with absolutely everyone rumble on this week, and this time it’s an update on his beef with Apple. The iPhone maker has asked a federal judge to dismiss the antitrust lawsuit levelled against them by the tech tycoon in August, calling it “speculation on top of speculation.” For those not in the know, Musk’s companies, xAI and X Corp, allege Apple’s deal with OpenAI limits competition for rival AI chatbots like Grok, but Apple counters that the agreement is “expressly not exclusive” and that it’s free to partner with other generative AI providers. In any case, they’re certainly not going to be “partnering” with xAI now, are they?
  

• Another big-name exec has got the chop at an enterprise-level business, demonstrating the growing trend to integrate AI systems at all levels – and apparently all costs, too. Starbucks’ Chief Technology Officer, Deb Hall Lefevre, has resigned amid a broader tech overhaul led by new CEO Brian Niccol, who is aiming to reverse six straight quarters of sales decline. Ningyu Chen, formerly Senior VP of Global Experience Technology, has been named interim CTO. The restructuring includes mass job cuts, store closures, and AI-driven initiatives (automated inventory, barista assistants, new point-of-sale systems) to boost efficiency and performance.
  

• You may have seen her – you simply may not care – but the poor fully AI-generated “actress” Tilly Norwood has come under fire week for well, not being human. Developed by AI studio Particle6/Xicoia, critics have called her rollout a “psy-op” designed to normalize generative AI in entertainment. Though agents have shown interest, Tilly is not human; her speech, movement, and presence are controlled by models overseen by humans. Critics warn that positioning avatars like her risk undermining real actors and their work.
  

  

  LoML, Tilly Norwood. Source: Particle6/Xicoia

🧩 Other Bits

• The nauseatingly cute “Hug my younger self” trend allows people to use AI for a bitta self-love. Powered by Google Gemini’s AI, people are creating nostalgic images of their current and childhood selves embracing, and predictably it has gone viral worldwide driving millions of app downloads. Sour pusses say it raises privacy concerns, but as someone who was never hugged as a child, I couldn’t care less.
  

  

  🥹 Source: Munsif Daily/Google Gemini

  
  

• One for identifying “that rash”: Delphi-2M can predict an individual’s risk of developing over 1,000 diseases, sometimes as far as 20 years ahead, using medical records and lifestyle data. Designed from a LLM Delphi-2M offers risk estimates rather than diagnoses, aiming to support preventive health monitoring.
  

• Gym enthusiasts and Peloton owners will soon be able to be berated by AI trainers, as the struggling fitness firm looks to kickstart growth. Peloton has revamped its entire product portfolio with AI-enhanced personalization features as part of a turnaround strategy to recover from declining subscriber growth.
  

• Another totally depressing insight into the human condition – and those willing to exploit it – CrushOn.AI is the latest app to join the burgeoning AI companion app market, with focus on romantic AI relationships. The emotional AI companion category has experienced 88% year-on-year growth in downloads, while the category as a whole has grown 60% since 2024.
  

  

  Source: We screen-shotted it, unfortunately/Google

  
  

• DoorDash has introduced a new creator program for short-form videos integrated into the platform. It adds new AI-powered features for personalization and discovery, meaning users can get dine-in rewards for signing up to the program. Singing for ya supper, literally.
  

• Google Cloud has announced an expanded AI infrastructure strategy focusing on startup partnerships to accelerate deployment of AI workloads and maintain competitive positioning. Schneider Electric and Nvidia, meanwhile, have signed a strategic partnership to develop AI data center infrastructure.
  

📈 Trending tools, models & apps this week

📋 LLM Leaderboard

📲 Trending tools & apps

• SaveClip lets you capture any link, video, or snippet from the web and store it beautifully. Think of it like a “save for later” tool that actually works. You can tag, categorize, and even AI-summarize your saves. Goodbye chaotic tabs, hello digital second brain.

• LangLime is your personal AI language lab. It helps you learn, translate, and chat in any language — with real conversational feedback, not flashcards. Whether you’re polishing Spanish or flirting in French, LangLime adapts to your tone and skill level.

• Casio Moflin is an AI pet with actual emotions. It learns from how you interact, responds with different moods, and even remembers you. It’s part robot, part pet therapy — less feeding, more feeling.

💸 Financials

• A strategic partnership between NVIDIA and OpenAI has been mooted, to deploy 10 gigawatts of NVIDIA systems by 2026. NVIDIA will invest up to $100B to power next-generation AI infrastructure for OpenAI's models.
  

• Cerebras Systems announced the completion of an oversubscribed $1.1B Series G funding round at an $8.1B post-money valuation, positioning the AI chip maker for continued growth in high-performance AI computing.
  

• Modular has raised $250M in its third financing round to continue building AI's unified compute layer – a hypervisor for AI that aims to simplify and accelerate AI infrastructure deployment.
  

• Lila Sciences raised a $235M Series A funding round, co-led by Braidwell and Collective Global, to develop its scientific superintelligence platform aimed at accelerating drug discovery and scientific research.
  

• CrowdStrike announced the acquisition of Pangea for $260M, integrating AI-powered detection and response capabilities into its cybersecurity platform to enhance threat detection and automated response.
  

• New York-based Flying Tulip, a full-stack on-chain exchange platform leveraging AI for trading infrastructure, announced it raised $200M to expand its blockchain-based trading capabilities.
  

• F5 Networks announced plans to acquire CalypsoAI for $180M, enhancing its security platform with advanced AI capabilities focused on protecting AI/ML models and applications in enterprise environments.
  

• Signal AI raised $164M to expand its AI-driven B2B analytics and intelligence platform, signaling robust investor interest in enterprise AI solutions that help businesses make data-driven decisions.
  

• San Jose-based Empower Semiconductor, world leader in powering AI-class processors, announced closing of more than $140M in Series D financing to scale production of energy-efficient power solutions for AI chips.
  

• Oracle is looking to secure $15B in funding to expand AI infrastructure capabilities despite rising interest rates, though analysts warn of execution risks in a competitive cloud market.
  

• Meta has acquired Rivos, a startup developing custom AI chips, to strengthen its AI hardware capabilities and reduce dependence on third-party chip manufacturers for AI infrastructure.
  

🕵️‍♂️ FREE ENTRY TO OUR INVITE-ONLY AI CHAT ON TELEGRAM…

If you share this newsletter with a friend and they actively sign up for the Big Machines newsletter, we will send you access to our invite-only Big Machines Telegram group, which is full of builders, investors, founders, and creators.

Access is now only granted to those who refer our newsletter to active subscribers, which means if you sign up on your work email, we will know you sneaky bastards.

This would kill our open rate, so please don't do that, we beg.

👋 Here’s a nice story to get you through the week… see you next time

Saved the best till last 😅…

And like that, that’s that. See ya next week.

Sam, Grant, Matt, Mike and the Big Machines team.