Top Stories

Behind the scenes of “AI slop” videos and their creators

Tech journalist Taylor Lorenz sits down with Washington Post journalist Drew Harwell, who has been diving deep into the world of "AI slop" videos, and the people that create them. It's an interesting look at its natural birth from contemporary meme culture and generation, the surprising reality of who these creators really are, and how they themselves see little future in it.

Despite AI replacing us all, Anthropic seeks to triple its international workforce

Anthropic says it will triple its international workforce and grow its applied AI team fivefold in 2025, as it chases enterprise customers outside the US. The company claims more than 300,000 enterprise customers and says nearly 80% of usage is now outside the US, with fresh offices in Tokyo and new hubs in Dublin, London and Zurich. The rapid expansion puts it in a head-to-head landgrab with OpenAI, Microsoft and Google, although some of the numbers quoted, like a $5 billion run-rate in under a year, invite a raised eyebrow.

• Rapid headcount expansion. Country leads are being hired across countries such as India, South Korea, and Singapore, with broader pushes across the UK and Europe. Anthropic’s new international MD, Chris Ciauri, is driving the rollout.

• Big growth claims. They say usage in South Korea, Australia and Singapore is outpacing the US, and that Claude Code is already a $500 million product after a 10x usage jump in three months.

• Enterprise wins and distribution. Case studies include Novo Nordisk, SK Telecom, NBIM and CBA, plus integrations via AWS Bedrock and Google Vertex to fit hybrid IT stacks.

The bottom line: The enterprise AI race is going global and Anthropic wants to be the pure-play alternative to cloud-suite incumbents. If even half these impact stats hold up, that is meaningful. Also ironic that AI bosses keep predicting mass redundancies while simultaneously announcing hiring sprees to sell more AI.

Insert yourself into videos with the Sora 2 app

OpenAI has launched Sora 2, a video and audio generation model they claim is a major leap in physics fidelity and controllability, plus a new invite-based Sora social app built around “cameos” that let users insert their likeness and voice into scenes. They say Sora 2 handles world state and failure more realistically, which means a missed basketball shot now rebounds instead of teleporting in. The app is rolling out in the US and Canada first, with free usage limits and a Pro tier coming.

• Big technical claims. OpenAI calls Sora 2 a “GPT‑3.5 moment for video” and says it is “better about obeying the laws of physics”. It supports multi-shot prompts, synchronised dialogue and sound effects, and styles from cinematic to anime.

• Social app and “upload yourself”. The Sora app focuses on remixing and cameos after a short identity capture, although real-world consent and deepfake risks will need more than a safety doc to settle.

• Safety and monetisation. OpenAI promises wellbeing tools, teen limits, and says it is not optimising for watch time. Monetisation is framed as paying for extra generations when compute is tight, which sounds benign but could shift later.

Why it matters: If Sora 2’s physics and control hold up, video gen becomes a more credible creative and prototyping tool. The bigger story is OpenAI moving from model to consumer platform, which raises trust, consent and distribution questions.

In Brief

• Engineers create first artificial neurons that can potentially communicate with living cells

• AMD’s guide to vibe-coding at home

• Try Opera’s new Neon AI browser

Market Trends

ChatGPT now has parental controls

OpenAI is testing a safety routing system in ChatGPT that detects emotionally sensitive conversations and flips the model mid-chat to what it calls GPT-5-thinking, alongside new parental controls for teen users. The move follows incidents where ChatGPT validated delusional thinking and a wrongful death lawsuit tied to a teen’s suicide. The new “safe completions” approach aims to answer sensitive questions safely rather than simply refuse, which OpenAI says will be refined over a 120-day iteration period.

• Safety routing and model swaps. Nick Turley, who leads the ChatGPT app, says routing happens per message and is temporary, and ChatGPT will reveal which model is active when asked. This comes after backlash when OpenAI made GPT-5 the default and users demanded access to the more agreeable GPT-4o.

• Parental controls and teen safeguards. Parents can set quiet hours, disable voice and memory, remove image generation, and opt out of model training. Teen accounts get stricter content filters and a system that flags possible self-harm, with human reviewers and alerts to parents, and potentially law enforcement if a life-threatening risk is detected.

• Mixed reactions and trade-offs. Some welcome tighter guardrails, others call it infantilising and say it degrades the service. OpenAI admits false positives will happen, arguing it is better to alert a parent than stay silent.

Why it matters: Safety layers are becoming core product features, which is as much about liability and trust as it is about ethics. If OpenAI can balance safety with usefulness, it sets a template others will follow.

Meta has acquired chip maker Rivos
Meta is said to be acquiring Rivos, a GPU startup, in a bid to accelerate its in‑house silicon and reduce reliance on Nvidia for AI workloads. The deal, reported via a person familiar with the matter, would bolster Meta’s Meta Training and Inference Accelerator programme and give it another shot at building competitive training and inference chips. Terms were not disclosed, although Rivos recently sought funding at a $2 billion valuation.

• GPU push and deal details. Rivos is working on its own GPU, which could complement or compete with Meta’s internal designs. Meta still spends billions on third‑party GPUs, so owning more of the stack could mean lower costs and better control of model schedules.

• Capex and infrastructure. Meta has made AI the top priority, with up to $72 billion in capex this year including AI infrastructure. It also raised $29 billion to build a massive Louisiana data centre, which shows this is not a side project.

• Pace and prior M&A attempts. Insiders say Meta’s custom chips have not progressed as fast as Mark Zuckerberg would like, although the company disputes this and says the work is moving quickly. Earlier talks to buy FuriosaAI for around $800 million fell through.

Why it matters: Big AI players want sovereignty over silicon which means cost, performance and supply chain independence. If Meta lands credible in‑house GPUs, it dents Nvidia’s grip and tightens Meta’s model roadmap.

Hollywood unions slam “AI actress” Tilly Norwood
Hollywood’s biggest actors’ union has slammed Tilly Norwood, an AI-generated “actress” from London startup Xicoia, calling her a threat to human performers. SAG-AFTRA says the character was trained on “countless professional performers” without consent or pay, and warned studios they must bargain before using synthetic performers. Creator Eline Van der Velden insists Tilly is a creative experiment, not a replacement, while also claiming that talent agents are already interested.

• Union pushback. SAG-AFTRA says Tilly “has no life experience to draw from, no emotion,” and argues audiences do not want content “untethered from the human experience.” The guild links the row to last year’s strike over AI likeness protections.

• Creator’s defence. Van der Velden frames Tilly as a new tool like animation or CGI, saying “she represents experimentation, not substitution.” Xicoia was spun out from Particle6 and unveiled Tilly at the Zurich Summit.

• Industry reaction. Emily Blunt called the development “really, really scary,” while Whoopi Goldberg said the model could mix traits from “5,000 other actors,” which sounds like an unfair advantage. Rumbles of interest from agents suggest the market will test that claim fast.

Why it matters: AI-generated talent is colliding with contracts, compensation and consent in plain sight. If agents start signing synthetic performers, expect a rapid tightening of union rules and legal tests around training data.

Hit reply to let us know which of these stories you found the most important or surprising! And, if you’ve stumbled across an interesting link/tweet/news story of your own, send it our way at [email protected] It might just end up in the next issue!

Thanks for reading. Stay tuned for the next AI Leadership Weekly!

Brought to you by Data Wave your AI and Data Team as a Subscription.
Work with seasoned technology leaders who have taken Startups to IPO and led large transformation programmes.