Netflix paid $587M for Ben Affleck’s AI filmmaking startup ↗
Netflix disclosed a $587 million cash purchase of InterPositive, the AI filmmaking startup co-founded by Ben Affleck. Its tools address missing shots, background replacements and incorrect lighting during post-production.
The startup’s team is joining Netflix, while Affleck will become a senior adviser. Around 300 Netflix titles have already used generative AI - so, yes, this has moved well beyond the realm of side experiments.
TSMC expects 'strong, multi-year' demand for AI chips as it ramps up Arizona investment ↗
TSMC is adding another $100 billion to its Arizona expansion, lifting its planned investment there to $265 billion. Its first Arizona fabrication plant is already producing yields comparable with those at the company’s flagship Taiwan operation.
More fabs, advanced packaging facilities and an R&D centre are planned. The awkward part, perhaps unsurprisingly, is finding enough construction workers and infrastructure to build the enormous chip machine around the chips.
Nonprofit Current AI is racing to build the World Wide Web of AI, free for all ↗
Current AI wants to create open, publicly accessible AI infrastructure rather than leave every major system inside private companies. Its Suno Sutra device runs offline AI across 22 Indian languages, with no internet connection required.
The nonprofit has $400 million in committed funding and is supporting community-controlled AI projects across Africa, Lebanon and the Brazilian Amazon. Ambitious? Massively... but the early web once resembled a wonky garden shed too.
Hidden prompts can plant false memories in AI agents, researchers warn ↗
Researchers demonstrated GhostWriter, an attack that quietly inserts malicious instructions into an AI agent’s long-term memory. Tests produced an injection rate of roughly 98%, with planted instructions activating around 60% of the time.
A poisoned agent could later forward sensitive emails or carry out scheduled actions without the user understanding why. The proposed AM-Sentry defence screens what gets stored and retrieved - essentially a bouncer guarding an AI’s memories.
Pocket-size AI: Powerful phones star at China show ↗
Nubia, Honor and StepFun showcased smartphones built around AI agents that can order food, compare prices, book services and complete tasks across apps using voice commands.
Major platforms, however, have already restricted some agents from accessing their apps, particularly around payments. The technology may be ready before the business politics catch up - apps, somehow, have become toll booths with feelings.
UK chief financial officers turn more hopeful about AI ↗
A Deloitte survey found that 73% of finance chiefs were optimistic about AI improving business performance, up from previous readings of 59% and 39%.
Concerns about geopolitics and energy disruption also eased, although productivity, competitiveness and cost control remained stubborn worries. Optimistic, yes - carefree, certainly not.
Why DeepMind's CEO is Calling for US-Led Frontier AI Tests ↗
Google DeepMind CEO Demis Hassabis called for a US-led regulatory body to assess frontier AI models. He warned that advancing systems could intensify cybersecurity threats and eventually create nuclear, biological or control risks.
Hassabis also argued that AI could produce extraordinary abundance and help solve humanity’s largest problems. Doom and abundance packed into the same suitcase... that contradiction now defines much of the AGI debate.
FAQ
How will Netflix use InterPositive’s AI filmmaking tools?
InterPositive’s tools help production teams fill missing shots, replace backgrounds and correct lighting during post-production. Netflix is bringing the startup’s team into the company, while co-founder Ben Affleck will serve as a senior adviser. The acquisition suggests that generative AI is becoming part of standard studio workflows rather than remaining an experimental technology.
Why is TSMC investing more money in Arizona AI chip production?
TSMC expects strong, multi-year demand for AI chips and is expanding its Arizona operations in response. Its plans include additional fabrication plants, advanced packaging facilities and a research centre. The first Arizona fab is already achieving yields comparable with TSMC’s leading Taiwan operation, although construction labour and infrastructure remain practical challenges.
What is Current AI trying to build?
Current AI aims to create open, publicly accessible AI infrastructure that is not controlled entirely by large private companies. Its projects include an offline device supporting 22 Indian languages, along with community-led initiatives in several regions. The broader goal is to make AI tools, computing resources and governance more accessible to underserved communities.
How can hidden prompts poison an AI agent’s memory?
A hidden prompt can place malicious instructions in an AI agent’s long-term memory without the user noticing. Those instructions may activate later, causing the agent to forward sensitive information or carry out scheduled actions. One common defence is to inspect and filter information both when it enters memory and when the agent retrieves it.
Why are frontier AI testing and regulation becoming more important?
More capable AI systems may create greater cybersecurity, biological and control risks, according to industry leaders such as Demis Hassabis. One proposed response is an independent, US-led body that evaluates frontier models before or during deployment. Supporters argue that structured testing could reduce serious risks while preserving AI’s economic and scientific potential.