The Weekly AI Digest
Week of 20–26 March 2026
Top 5 AI Stories This Week
OpenAI Shuts Down Sora, Disney Drops $1 Billion Investment
OpenAI announced on March 24 that it is discontinuing Sora, its AI video generation app, just six months after launch. The iOS app, API and Sora.com will all be wound down. OpenAI cited the need to prioritise compute for its core enterprise products, with video generation proving far more resource-intensive than text. The Sora research team will pivot to world simulation research for robotics. Disney ended its partnership with OpenAI, which had included plans for a $1 billion stake. No money ever changed hands. Meanwhile, Elon Musk's xAI moved to fill the gap, declaring the next Grok Imagine release will be a major upgrade to its video generation capabilities.
Why it matters: OpenAI's decision signals that even the largest AI labs face hard trade-offs on compute allocation. Video generation remains expensive and difficult to monetise — Sora earned just $2.1 million in its lifetime. Businesses building on AI video tools should consider platform risk carefully and evaluate alternatives like xAI, Runway and Pika.
Visa Launches Agentic Ready Program Across Europe
Visa announced its Agentic Ready program on March 17, enrolling 21 major banks including Barclays, HSBC UK, Revolut, Banco Santander, Commerzbank and Nationwide Building Society. The program gives financial institutions a structured way to test payments initiated by AI agents on behalf of consumers before they encounter them at scale. Santander and Visa also completed Latin America's first end-to-end agentic commerce transactions across five markets. Visa's Crypto Labs unveiled a command-line interface allowing software agents to initiate payments directly from a terminal.
Why it matters: AI agents that can browse, decide and buy are moving from concept to infrastructure. Visa's involvement signals that the payments industry is taking agentic commerce seriously enough to build dedicated rails for it. Businesses should start thinking about how their products and services will be discovered and purchased by AI agents, not just humans.
Super Micro Co-Founder Arrested for Smuggling $2.5B in AI Chips to China
Three people affiliated with server maker Supermicro, including co-founder Wally Liaw, were charged by federal prosecutors for conspiring to smuggle $2.5 billion worth of Nvidia AI chips to China in violation of US export controls. Liaw and a second defendant were arrested; a third remains a fugitive. Prosecutors allege the group sold servers to a front company in Southeast Asia, which repackaged and forwarded $510 million worth of banned hardware to China. They reportedly used hair dryers to remove labels and serial numbers from real machines and placed them on decoys. Supermicro's stock plunged 33% on the news.
Why it matters: This is the largest AI chip smuggling case to date and a stark warning about enforcement of export controls. Companies in the AI hardware supply chain face increasing scrutiny. The case also highlights how valuable advanced AI chips have become — worth building elaborate international smuggling operations around.
Google Signals Ads Are Coming to Gemini
Google confirmed it is no longer ruling out advertising in its Gemini AI assistant, with AI Mode serving as the testing ground for ad formats that could eventually expand across the company's AI products. Gemini now has 750 million monthly active users, up from 350 million a year ago. The company also expanded its Personal Intelligence feature to all US users, connecting Gmail, Photos and Calendar data to provide more personalised responses. When asked whether personal data could influence search ads, Google's Nick Fox said it was 'TBD' but that targeting could be 'contextually consistent.'
Why it matters: Google monetising Gemini with ads would reshape the AI assistant landscape. For advertisers, it opens a potentially massive new channel. For businesses relying on AI assistants to surface information neutrally, it raises questions about whether AI recommendations will increasingly be influenced by ad spend rather than relevance.
Britannica and Merriam-Webster Sue OpenAI Over RAG Copyright
Encyclopedia Britannica and Merriam-Webster filed a federal lawsuit against OpenAI on March 13, alleging the company scraped nearly 100,000 copyrighted articles to train its models without permission. The lawsuit goes further than previous copyright cases by explicitly targeting OpenAI's RAG (retrieval-augmented generation) system, arguing it reproduces copyrighted content when generating real-time responses. Britannica also invoked the Lanham Act, alleging that when ChatGPT hallucinates false information and attributes it to Britannica, it constitutes trademark violation.
Why it matters: This lawsuit could set precedent for how RAG systems interact with copyrighted content — a mechanism most enterprise AI deployments rely on. If courts rule that RAG constitutes copyright infringement, it would force a fundamental rethink of how AI systems access and present third-party information. Enterprise customers should monitor this case closely.
Australia Watch
Deloitte Warns of 'Sliding Doors' Moment on AI
Deloitte Access Economics released a report warning that Australia has a narrow window to position itself as a regional AI infrastructure hub but risks falling behind unless investment and policy action accelerate. The firm estimates Australia needs approximately A$52 billion ($36 billion) in digital infrastructure investment by 2030 to capture the opportunity.
The window for Australia to establish itself as an AI hub is narrowing as other countries accelerate investment.
New Rules for AI and Data Centres
The Albanese government announced a strict five-step framework that global tech companies must follow to build data centres or operate AI systems in Australia. The rules come as the government prepares to establish an AI safety institute in the first half of 2026, following the Canada-Australia AI Safety MOU announced earlier this month.
Increased regulatory requirements may slow deployment but signal Australia's intent to maintain oversight of AI infrastructure.
AI-Designed Cancer Vaccine Shrinks Dog's Tumour by 75%
Australian AI consultant Paul Conyngham used ChatGPT, AlphaFold and Grok to design a personalised mRNA cancer vaccine for his rescue dog Rosie. UNSW's RNA Institute manufactured the vaccine in under two months. The tumour on Rosie's leg shrank by 75% within a month of the first injection — the first personalised cancer vaccine ever designed for a dog.
A remarkable demonstration of AI tools being used to accelerate personalised medicine, even outside traditional research settings.
Emerging Trends
Agentic Commerce Is Getting Real Infrastructure
Visa's Agentic Ready program, with 21 banks enrolled and live transactions in Latin America, marks the moment agentic commerce moved from demos to payment rails. When the payments industry builds dedicated infrastructure for AI agents, it signals that agent-to-business transactions are no longer hypothetical.
Businesses should start considering how they'll authenticate, authorise and interact with AI agents as customers.
The Compute Trade-Off Is Sharpening
OpenAI shutting down Sora to redirect compute to enterprise products illustrates a broader industry dynamic: AI labs are being forced to choose where to allocate scarce GPU resources. Expect more experimental products to be cut as labs focus on revenue-generating enterprise use cases.
The survivors will be the products that justify their compute costs.
Copyright Law Is Catching Up to RAG
The Britannica lawsuit explicitly targeting RAG — not just training data — opens a new legal front. Most enterprise AI deployments use some form of retrieval-augmented generation. If courts rule that RAG reproduces copyrighted content, it could reshape how businesses deploy AI systems that draw on external information sources.
Legal teams should be reviewing their AI procurement agreements now.