AI News – March 29, 2026: The End of Sora, the Rise of Hyperagents, and the Reality Check on AI Use

The week from March 23 to 29, 2026 will be remembered as the great clash between generative hype and market reality. While OpenAI quietly freezes the expensive

If last week we analyzed the triumph of hardware infrastructure with Nvidia's AI Factories and Cerebras chips, the days from March 23 to 29, 2026 mark a turning point for software and the public perception of technology.

It was the week of major "reality checks." While vertical startups raise monstrous capital, once-celebrated models like OpenAI Sora are quietly being shelved. Meanwhile, Pew Research's demographic data reveals that, despite the newspaper headlines, Americans still don't trust AI to read the news. On the medical front, however, the algorithm literally saves lives, officially entering hospital wards with FDA approval.

Here are the 5 key news stories of the week, analyzed to understand their real impact.

1. DGM-Hyperagents: Welcome to the Era of AI that "Actually Does Your Job"

Chatbots are outdated. The industry is massively shifting towards agents capable of autonomously operating on complex tasks.

🔍 What happened: As reported in the technical deep dive by DevFlokers on the March 23-24 releases, the market was shaken by the launch of DGM-Hyperagents. These are next-generation agentic architectures designed not to "answer questions," but to execute multi-step workflows in total autonomy. Alongside these, there is the consolidation of extremely efficient models like Qwen 3.5 Small and the new OpenClaw interfaces.

💡 Why it matters: It's the definitive shift from "Generative AI" to "Agentic AI." A Hyperagent doesn't write you a draft email; it analyzes your CRM, identifies at-risk customers, formulates a personalized offer, sends the emails, reads the responses, and updates the database, all in the background.

🎯 Our take: As we anticipated in our special on How Artificial Intelligence supports the supply chain and management, companies no longer want "oracles" to chat with, but "digital colleagues" who complete operational processes. Hyperagents are the answer to this need.

2. The End of Sora and the Triumph of Vertical AI: The Amity Case

While generalist models show the first signs of infrastructural weakness, capital is pouring into vertical and hyper-specialized solutions.

🔍 What happened: The daily summary by LabLA on March 25 confirmed a rumor circulating for months: OpenAI has quietly shut down the Sora project, its revolutionary (and extremely expensive) text-to-video generator. Unsustainable inference costs, combined with insurmountable safety and copyright issues, pushed the company to freeze the tool. Simultaneously, Amity, a Vertical AI platform specialized in the legal and fintech sector, closed a Series B funding round of a whopping $100 million.

💡 Why it matters: The "graveyard" of Sora demonstrates that the wow effect does not guarantee a sustainable business model. AI-generated videos are fascinating, but cost too much in terms of computing power (GPUs) compared to the value they return. In contrast, "Vertical" Artificial Intelligence (trained on specific data from a single sector, like Amity for Fintech) solves boring but extremely profitable problems, attracting big capital.

🎯 Our take: This is the bursting of the "jack-of-all-trades foundation models" bubble. To understand how this transition will change the way designers and videomakers work, freed from the short-term threat of entirely AI-generated videos, we refer you to our in-depth analysis on Artificial Intelligence and creative work: what changes in the future?.

3. Infrastructure and WWDC 2026: Siri Becomes Intelligent and Nvidia Opens Up to Kubernetes

The ecosystem consolidates: Apple prepares its voice revolution, while Nvidia and Hugging Face provide the tools to evaluate and scale these technologies.

🔍 What happened: The daily video recaps by AI Pulse and AI News Today revealed a trio of crucial announcements on the infrastructure front:

  1. Nvidia released open-source versions of its GPU orchestration tools for Kubernetes, making it easier for companies to manage their clusters.
  2. Hugging Face launched a new independent framework to objectively evaluate the performance of Voice Agents.
  3. Details of the imminent Apple WWDC 2026 have leaked: Siri will receive its biggest update ever, integrating an edge-based version of Gemini and the new "Apple Intelligence" models to act directly on iPhone apps.

💡 Why it matters: Voice AI is the next battlefield. To understand if a voice agent is truly useful (and not just a robotic voice reading Wikipedia), rigorous benchmarks like those proposed by Hugging Face are needed. Meanwhile, the new Siri promises to bring Contextual Intelligence to the pockets of a billion users.

🎯 Our take: The infrastructure is finally maturing. As highlighted in our guides on wearable devices and contextual intelligence, the future of human-machine interaction will be off-screen and voice-driven.

4. AI in the Hospital Ward and the Shadows of Algorithmic Psychiatry

The medical sector shows us the profound dichotomy of AI: a life-saving tool when used by professionals, a potential danger if left uncontrolled in the hands of the young.

🔍 What happened:

  • US News dedicated an exclusive Q&A to the Viz Hemorrhage system, an FDA-approved AI that analyzes brain scans. The system, recently awarded the prestigious Edison Award 2026, can detect micro-brain hemorrhages with speed and precision superior to the human eye, seamlessly integrating into the hospital workflow.
  • As a counterpoint, a UConn Health Minute bulletin raised an alarm about the growing use of chatbots among teenagers for mental health. Child psychiatrists report serious limitations: kids use commercial LLMs as if they were therapists, exposing themselves to misdiagnoses and dangerous emotional isolation.

💡 Why it matters: Viz Hemorrhage demonstrates that AI succeeds when it is "narrow" and supervised (Human-in-the-loop). The UConn alarm, however, exposes the dramatic risks of delegating psychological care to software designed to simulate empathy but lacking clinical understanding.

🎯 Our take: Regulations must distinguish between Software as a Medical Device (regulated and life-saving) and consumer chatbots masquerading as confidants. We explored this psychological emergency in our recent article on Addiction to AI interfaces and emerging relational risks.

5. Reality Check: Only 1% of Americans Use AI for Breaking News

Despite fears that Artificial Intelligence would destroy publishing and supplant Google in real-time news search, the data tells a very different story.

🔍 What happened: An authoritative survey by Pew Research, picked up and analyzed by MediaPost, revealed that only 1% of adults in the United States use AI chatbots (like ChatGPT or Claude) to get information on "breaking news". The public continues to strongly prefer traditional search engines, direct news outlets, or social media for real-time events.

💡 Why it matters: It's a colossal scaling back of Silicon Valley's projections. It shows that the public is deeply aware of the "hallucination" problem (facts invented by AI). When it comes to news, politics, or ongoing catastrophic events, trust in generative algorithms is close to zero.

🎯 Our take: Trust is built over decades but destroyed in an instant. As analyzed in our special on Fake News and AI: An Information War, chatbots are excellent at summarizing existing documents but are terrible reporters. Traditional publishing still has an unbridgeable competitive advantage: human editorial responsibility.

FAQ: Frequently Asked Questions of the Week

1. Why did OpenAI shut down the Sora project? Although OpenAI has not released official statements of technical failure, analysts agree that the processing costs to generate photorealistic videos were economically unsustainable at scale. Add to this the enormous legal pressures for copyright infringement (the unauthorized use of protected videos for training) and the inability to guarantee robust safety filters against the creation of political-pornographic deepfakes ahead of the imminent election cycles.

2. What are DGM-Hyperagents? DGM (Dynamic Goal Management) Hyperagents are next-generation AI software. Unlike old agents that stopped at an obstacle, a Hyperagent can dynamically remodulate its intermediate goals if it encounters an unforeseen event, completing complex work tasks (e.g., "Extract the data, compile the tax report, and send it to the directors") with a success rate close to that of a human.

3. What is the difference between General AI and Vertical AI? General AI (like ChatGPT) is trained on all human knowledge and can write a poem, translate code, or give you a tiramisu recipe. It is versatile but superficial. Vertical AI (like Amity) is trained exclusively on data from a specific sector (e.g., jurisprudence or finance). It can't write poems, but it can review a corporate merger contract with the precision of a senior lawyer.

4. What is Viz Hemorrhage and why did it win the Edison Award? It is an FDA-authorized medical software based on Machine Learning. It connects to CT machines in hospitals and analyzes images in real-time. If it detects a suspected brain hemorrhage, it sends an immediate alert to the on-call neurosurgeon's smartphone, drastically reducing intervention times (where every minute lost equals millions of dead neurons). It won the Edison Award for its practical, life-saving impact.

5. Why do people avoid using AI for breaking news? For two structural reasons: latency and hallucinations. Large Language Models (LLMs) have a "knowledge cutoff" date and struggle to reliably index events happening at this very moment. Furthermore, the AI's tendency to "invent" plausible but false details (hallucinations) to please the user makes it an unreliable source for information where factual accuracy is a matter of life or death.

Bibliographic References and Sources