March 26, 2026 – AI Daily Recap

The AI industry just served up one of its most eventful days in recent memory. OpenAI pulled the plug on Sora and torpedoed a billion-dollar Disney deal, Arm shipped its first-ever physical chip, Jensen Huang declared AGI has arrived, Anthropic gave Claude the keys to your Mac, a major supply chain attack rattled the open-source ecosystem, and AI2 dropped an open-source browser agent that punches well above its weight. Here’s what you need to know.

OpenAI Shuts Down Sora, Disney Walks Away from $1B Deal

In a move that stunned Hollywood and Silicon Valley alike, OpenAI announced it is discontinuing its Sora video generation app just six months after launch. The shutdown immediately killed Disney’s planned $1 billion investment in OpenAI, which had been built around the Sora platform and access to more than 200 Disney, Marvel, Pixar, and Star Wars characters. According to reports, Disney teams were informed that Sora was being killed just 30 minutes after their last working meeting with OpenAI. The decision appears tied to OpenAI’s looming IPO — Sora’s massive GPU consumption and unproven monetization made it a liability on the balance sheet. OpenAI is now redirecting those resources toward its next major model, codenamed “Spud,” and broader AGI infrastructure buildout.

Arm Ships Its First Chip in 35 Years — and Meta Is the Launch Customer

In a historic pivot, Arm launched the AGI CPU, the company’s first physical chip in its 35-year history. The 136-core, 3nm data center processor is designed specifically for AI inference workloads, and Arm claims it delivers twice the performance-per-watt of comparable x86 racks. Meta signed on as the lead launch customer, co-developing the chip to optimize gigawatt-scale infrastructure for its family of apps and to work alongside Meta’s own custom MTIA accelerators. Other early partners include OpenAI, Cerebras, Cloudflare, SAP, and SK Telecom. Up to 64 of the new CPUs — roughly 8,700 cores — can fit in a single air-cooled rack. Arm’s stock surged 16% on the news as analysts projected a significant revenue windfall from the shift to selling physical silicon.

Jensen Huang Declares “We Have Achieved AGI”

NVIDIA CEO Jensen Huang made waves with his bold claim that artificial general intelligence has been achieved — walking back his own 2024 prediction that AGI was still five years away. During an interview released on March 23, Huang defined AGI pragmatically as AI systems capable of autonomously building and operating billion-dollar businesses. The statement immediately sparked fierce debate across the AI community: critics questioned his narrow, commercially convenient definition of AGI, while supporters called it a pragmatic acknowledgment of how far the technology has come. At GTC 2026, Huang also projected $1 trillion in chip demand through 2027 and spotlighted new agentic AI tools as the next frontier.

Anthropic’s Claude Gains Full Computer Control on Mac

Anthropic rolled out computer use capabilities for Claude Code and Claude Cowork, letting the AI assistant point, click, and navigate your Mac to complete tasks autonomously. When Claude lacks the right tools for a job, it can now open files, use the browser, and run developer tools on its own. The feature is available to Claude Pro and Max subscribers on macOS, and pairs with Dispatch, the recently launched feature that lets users assign tasks from an iPhone and return to finished work on their desktop. Anthropic emphasized that Claude always requests permission before accessing new apps, and the capability remains in research preview as the company gathers feedback. This puts Anthropic squarely in the agentic AI race alongside OpenAI’s Operator and Google’s Project Mariner.

LiteLLM Supply Chain Attack Compromises Millions of Environments

A major supply chain attack hit LiteLLM, the widely used open-source AI gateway library with 97 million monthly downloads that’s present in 36% of all cloud environments. A threat actor known as TeamPCP published backdoored versions (v1.82.7 and v1.82.8) after stealing PyPI credentials through a compromised Trivy GitHub Action in the CI/CD pipeline. The malicious code harvested SSH keys, cloud credentials, and environment variables, then attempted lateral movement across Kubernetes clusters. The tainted packages were live for approximately three hours before PyPI quarantined them. Security researchers are now urging anyone who installed the affected versions to immediately rotate all credentials that were exposed on those systems.

AI2 Drops MolmoWeb, an Open-Source Browser Agent That Beats GPT-4o

The Allen Institute for AI released MolmoWeb, a fully open-source web agent that can navigate and complete tasks in a browser by interpreting screenshots the way a human would — clicking, typing, and scrolling without relying on underlying page code. Available in 4B and 8B parameter versions, the 8B model outperforms agents built on much larger proprietary models including GPT-4o on key web navigation benchmarks. Alongside the model, AI2 released MolmoWebMix, the largest publicly available dataset of human web-task execution ever assembled — featuring 30,000 human task trajectories across 1,100+ websites and 2.2 million screenshot question-answer pairs. Both the model and data are available on Hugging Face and GitHub.

The Big Picture

Today’s news hammers home just how fast the AI landscape is shifting. OpenAI’s Sora shutdown shows that even well-funded bets can be abandoned overnight when the math doesn’t work for an IPO. Meanwhile, the infrastructure race is heating up — Arm’s leap from IP licensor to chip maker and Jensen Huang’s AGI declaration both signal that the battle for AI compute is entering a new phase. On the product side, Anthropic giving Claude direct computer control represents a major step in the agentic AI push that defined GTC 2026. But the LiteLLM attack is a sobering reminder that as AI tooling proliferates, the software supply chain becomes an increasingly attractive target. And AI2’s MolmoWeb proves that the open-source community can still punch above its weight — sometimes literally outperforming the giants.

Sources

• Variety — OpenAI Will Shut Down Sora Video App; Disney Drops Plans for $1 Billion Investment
• The Hollywood Reporter — The Sora-Disney Collapse: What Does It Mean?
• CNBC — Arm Launches Its Own CPU, With Meta as First Customer
• TechCrunch — Arm Is Releasing the First In-House Chip in Its 35-Year History
• CNBC — Arm Jumps 16% as Company Expects Revenue Windfall from New Chip
• Yahoo Finance — Nvidia CEO Jensen Huang Claims AGI Has Been Achieved
• IBTimes — Jensen Huang Declares ‘We Have Achieved AGI,’ Sparking Debate
• CNBC — Anthropic Says Claude Can Now Use Your Computer to Finish Tasks
• Engadget — Claude Code and Cowork Can Now Use Your Computer
• The Hacker News — TeamPCP Backdoors LiteLLM Versions via Trivy CI/CD Compromise
• LiteLLM — Security Update: Suspected Supply Chain Incident
• AI2 — MolmoWeb: An Open Agent for Automating Web Tasks
• SiliconANGLE — AI2 Releases Open-Source Visual AI Agent
• Tech Startups — Top Tech News Today, March 25, 2026