📰 Welcome back!

If you needed an indicator of how quickly AI is evolving, you’ll understand what a difference a week can make.

OpenAI’s Sora 2’s launch has been seismic in many ways, its reverberations felt by way of copycats, deepfakes and general tomfoolery so imperceptible the panic button is being hit by leaders globally.

It’s probably of little surprise then, this week news is heavy on what exactly the global fun police want to do about it. Spoiler: they can’t decide.

Reactions will likely be glacial and the evolution of the tech will continue to outstrip measures at an alarming pace. We’re just here for the ride in the meantime.

🚀 What we’re covering today…

• 🤖 Pentagon’s AI apocalypse fears

• 🎥 Sora 2 copyright headache

• 🏛 Congress slows-down AI legislation

• 🇮🇹 Italy introduces AI law

• 🕵️ OpenAI closes “Chinese spy accounts”

• 🐈‍⬛ Intel unveils Panther Lake

• 👩‍⚖️ US judges AI blunders

• 🔗 eBay gives offenders a second chance

• 🕊️ Sonnet 4 is available in Xcode 26

• 🫧 Are we in an AI bubble?

🔴 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

• It might sound like a soft-porno spin on Terminator, but the Politico article “Killer Robots, AI Psychosis and Nuclear War: The Pentagon’s Biggest AI Fears” may be one of the more cheery AI reads you indulge in this week. This Q&A with a former Pentagon insider portrays the top echelons of the US’s intelligence services grappling with the existential risks posed by advanced AI. The article spotlights fears over fully autonomous weapons – “killer robots” – that act without human oversight; anxieties around “AI psychosis,”; and simulations revealing AI-initiated nuclear strikes, amplifying concerns that automation could lower the threshold for nuclear war.

  

  “Skull Net Rising” Source: We made it w/ Nano Banana

  
  

• When OpenAI launched Sora 2, in their infinite wisdom they did so without an opt-out copyright policy, which allowed users to generate videos featuring copyrighted characters unless rights holders explicitly excluded themselves. The policy took a shit in the bed, with users creating expectedly heinous* results (*see: fucking hilarious) such as Nazi SpongeBob and criminal versions of popular IPs. In response, OpenAI reversed course – now requiring positive permission (opt-in) from rights holders. Who’d have thunk it?
  

• The US federal government and its state legislators have continued to flip-flop on how to slow the AI doomsday clock in the past week. The fed’s shutdown is hampering Congress’s ability to advance AI legislation, effectively delaying a coordinated national policy framework. Key bills, such as an AI sandbox have stalled. Experts warn that when Congress resumes, priority will go to the National Defense Authorization Act (NDAA), meaning more bombs for foreign countries, less bandwidth for AI issues. As is time memoriam, a broad lack of federal oversight could lead to a more yankee-doodle-dandy, state-by-state patchwork of legislation.
  

• In a more remarkable turn of events, the Italian state has become a European forerunner and first EU member state to enact a national AI law aligned with the EU AI Act. Key features include mandatory human oversight, traceability of AI decisions, parental consent for users under age 14, criminal penalties (1-5 years) for harmful AI misuse (e.g. deepfakes or fraud), and rules over copyright and data mining. National authorities AgID (Digital Italy) and ACN (Cybersecurity Agency) will enforce it. Do ChatGPT: “Does carbonara include cream?” Do not ChatGPT: “Meloni covered in carbonara”.
  

  

  “Mamma Mia!” Source: We made it w/ Nano Banana

  
  

• OpenAI has banned multiple accounts linked to Chinese surveillance efforts, citing national security and misuse concerns. The company said the accounts were attempting to access its AI models to support state surveillance programs. This move reflects growing geopolitical tensions over AI access and control, especially between the US and China. It’s also a widening reflection of roles like OpenAI’s in global security dynamics.
  

• Intel has unveiled Panther Lake, its next-generation client processor line and the company’s first AI-centric PC platform, built on the Intel 18A process node. The architecture is already in production and is set to ramp to high-volume manufacturing later this year at Intel’s Fab 52 in Arizona. Panther Lake aims to combine performance and power efficiency improvements, targeting AI PCs, edge devices, and gaming systems. Alongside it, Intel previewed Clearwater Forest (Xeon 6+), its first server chip on 18A, scheduled for launch in the first half of 2026.
  

• US Senator Chuck Grassley has asked two federal judges – Julien Neals (New Jersey) and Henry Wingate (Mississippi) – to give them the skinny on whether genAI or automated tools played a role in drafting opinions the judges later withdrew due to serious factual errors. The inquiry reflects concerns AI misuse may undermine judicial integrity and transparency – certainly higher stakes than using it to write an AI letter not written by AI. Ahem.
  

  

  “Is this an AI, fellow kids?” Source: Getty Images

  
  

• eBay is launching AI Activate, a £3 million (US$4M) initiative in partnership with OpenAI. It will give 10,000 UK sellers free access to ChatGPT Enterprise and training. The programme offers up to 12 months’ use of advanced AI tools, including data privacy safeguards and higher usage limits than consumer accounts. Designed to help small businesses boost productivity in tasks like marketing, financial analysis, and business research, eBay also plans to help participants build custom GPTs tailored to their workflows.
  

• One that flew slightly under the radar but, Anthropic has announced that Claude Sonnet 4 is now generally available in Xcode 26, integrating directly into Apple’s IDE to support coding workflows. Developers can link their Claude accounts to Xcode and use natural-language prompts to debug, refactor, generate documentation, create SwiftUI previews, and make inline code changes.The feature is available to users on Pro, Max, Team, and Enterprise Claude plans.
  

• Are we all living in an AI bubble? Startups are now receiving a “disproportionate share of venture capital” – raising $73.1B in Q1 2025, or about 58% of all global VC funding – which has stirred warnings from top investors about excessively inflated early-stage valuations. Critics argue that many firms are being valued on hype rather than performance, with some commanding price multiples that vastly outpace their revenues, fuelling fears of a speculative bubble.

🧩 Other Bits

• Further up we discussed how a state vs federal approach to AI regulation is taking shape – this story is it in a microcosm: Wisconsin Governor Tony Evers signed a law (2025 Act 23) that expands the state’s criminal code to address AI misuse by “prohibiting the creation or distribution of intimate images generated without consent, especially if intended to harass or intimidate”.
  

• India’s Finance Minister Nirmala Sitharaman expressed concern over misuse of AI, revealing she’s seen deepfake videos of herself circulated online aimed at misleading citizens. “The Finance Minister called for collaboration among fintech innovators, investors and regulators to build a secure, inclusive and resilient tech system.”
  

• After Sora’s launch, the Apple App Store was flooded with “Sora”-branded copycat apps, many using names like “Sora 2” to ride search visibility.Some of these impostor apps slipped through Apple’s dragnet, and garnered tens of thousands of installs. Yet while many were later removed, a few remain live.
  
  

📈 Trending tools, models & apps this week

📋 LLM Leaderboard

📲 Trending tools & apps

• Sora 2 – for those who don’t yet have access, this Hugging Face space lets you try out OpenAI’s cinematic video model right now. Upload a prompt, pick your parameters, and watch it bring your imagination to life. It’s not official access, but it’s the next best thing.

• Vidau turns your ideas into studio-quality videos with zero editing skills required. Just describe your scene, choose a style, and let it produce cinematic clips instantly. Think of it as your AI video co-director.

• AI Comic Factory lets you create entire comics from text prompts. Write a story, pick a style, and watch it turn into a full illustrated strip — speech bubbles and all. Perfect for storytellers, meme-makers, and creative chaos.
  

💸 Financials

• AI startups secured over $965M in early-stage funding during the week of September 28-October 4, 2025, with Series A deals exceeding $343M and seed-stage investments totaling $602M, demonstrating continued investor confidence.
  

• Nvidia-backed Reflection AI raised $2 billion in funding, boosting its valuation to $8B. Founded by former Google DeepMind researchers, the company aims to challenge OpenAI with an open-source frontier AI lab approach.
  

• Periodic Labs raised $300M in seed funding led by Andreessen Horowitz. The startup's core bet is to develop AI platforms for materials science, using ML to simulate and predict material properties for research and manufacturing.
  

• Empower Semiconductor completed a $140M Series D funding to develop energy-efficient power management solutions for AI processors, addressing the growing demand for sustainable AI infrastructure.
  

• Dual Entry secured $90M to expand its AI-based accounting software for mid-market enterprises, aiming to automate financial workflows and reduce manual errors in accounting processes.
  

• Axiom Math AI raised $64M to develop self-improving AI systems for enterprise-level mathematical problem solving and IT applications, emphasizing continuous learning and optimization capabilities.
  

• Y Combinator-backed David AI secured $50M in funding to create diverse audio datasets for next-generation AI applications, addressing the critical need for high-quality training data.
  

🕵️‍♂️ 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.

The use of LLMs has made its way to Instagram and TikTok. We will never get bored of tricking family members into thinking there is a provocative trady or cleaner in the gaff. How many missed calls?

mmmjoemele

4M followers

View more on Instagram

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