AI News Week of March 27, 2026

AI News Week of March 27, 2026

Ryan Wong March 27, 2026 AI, News, Technology, Updates, OpenAI, Sora, Shopify, Anthropic, Claude, MCP, Google, Gemini, ByteDance, LiteLLM, Security

AI News Of The Week (27th March, 2026)

TLDR: Key AI Developments This Week

OpenAI shut down Sora standalone app and API, dissolving the $1 billion Disney licensing deal just six months after launch. Shopify activated agentic storefronts by default for all merchants across ChatGPT, Copilot, and Gemini with no setup required. Anthropic launched computer use for Claude with full mouse, keyboard, and screen control paired with Dispatch for task handoff. The Model Context Protocol reached 97 million monthly SDK downloads, becoming the de facto standard for AI agent tooling. A LiteLLM supply chain attack exposed SSH keys and cloud credentials across millions of users. ByteDance launched Dreamina Seedance 2.0 inside CapCut for mass-market video AI, and Google released Gemini 3.1 Flash Live for sub-second multimodal responses.

OpenAI Shuts Down Sora and the $1 Billion Disney Deal Collapses With It

On March 24, OpenAI announced it is discontinuing Sora — the standalone app, Sora.com, and the developer API — just six months after the app launched to the top of the App Store. The economics had been unsustainable from the start: inference costs were estimated at around $15 million per day at peak, against $2.1 million in total lifetime in-app revenue. Downloads peaked at 3.3 million in November 2025 and had already fallen 66% to 1.1 million by February. OpenAI's stated rationale is redirecting compute toward robotics and world simulation research. The closure also dissolves the $1 billion Disney investment and three-year licensing deal, under which Disney had agreed to let Sora generate videos featuring 200-plus characters from Disney, Marvel, Pixar, and Star Wars. The Sora 2 model itself remains available inside ChatGPT for Plus and Pro subscribers, but the standalone product is gone. Read more

Shopify Agentic Storefronts Go Live by Default for All Merchants Inside ChatGPT, Copilot, and Gemini

Starting March 24, Shopify activated Agentic Storefronts by default for all eligible US merchants, making their products instantly discoverable inside ChatGPT, Microsoft Copilot, Google AI Mode, and the Gemini app — with no setup required. Inventory, pricing, and product data stay synced centrally from the Shopify Admin. A new Agentic Plan also launched alongside, letting brands on any e-commerce platform add products to Shopify Catalog and sell across those same AI channels without migrating to Shopify. The move is underpinned by two open standards: the Agentic Commerce Protocol co-built with OpenAI and Stripe, and the Universal Commerce Protocol co-developed with Google and endorsed by more than 20 retailers including Walmart, Target, Visa, and Mastercard. AI-attributed orders on Shopify have grown 11x since January 2025. Read more

Anthropic Gives Claude Full Computer Control — Mouse, Keyboard, and Screen — Paired with Dispatch

On March 24, Anthropic launched Computer Use inside Claude Cowork and Claude Code, letting the AI take direct control of a Mac — opening apps, navigating browsers, filling spreadsheets, and running developer tools — using a three-tier hierarchy: connectors first, then browser control, then full screen interaction as a last resort. The feature pairs with Dispatch, Anthropic's phone-to-desktop task handoff tool, letting users assign work from their phone and return to completed results on their desktop. Claude always requests permission before accessing a new application and maintains blocklists for sensitive categories including finance and facial image scraping. The capability is available as a research preview for Pro and Max subscribers on macOS. Claude Code also received Auto Mode this week — an AI safety classifier that automatically approves routine developer actions while blocking destructive operations. Read more

MCP Hits 97 Million Monthly SDK Downloads, Confirmed as the De Facto Standard for AI Agent Tooling

Anthropic confirmed this week that the Model Context Protocol has reached 97 million monthly SDK downloads — up from approximately 2 million at launch in November 2024, a 4,750% increase in 16 months. The MCP server ecosystem now spans more than 5,800 community and enterprise servers covering databases, CRMs, cloud providers, productivity tools, and e-commerce platforms. For context, the React npm package took roughly three years to reach 100 million monthly downloads; MCP achieved comparable scale in 16 months. Every major AI provider — OpenAI, Google, Microsoft, Meta, and Amazon — now ships MCP-compatible tooling. Anthropic has donated MCP governance to the Agentic AI Foundation under the Linux Foundation, with OpenAI, Google, AWS, and Microsoft as founding members. Read more

LiteLLM Supply Chain Attack on PyPI Exposed SSH Keys and Cloud Credentials Across Millions of Users

A supply chain attack was discovered in LiteLLM v1.82.8 on PyPI this week: a malicious file embedded in the package silently harvested SSH keys, cloud credentials, and environment secrets on every Python startup, then attempted lateral movement across Kubernetes clusters. Andrej Karpathy amplified the warning, flagging the library's 97 million monthly downloads and noting that transitive dependencies mean the blast radius extends far beyond direct users. The attack is the second high-profile AI supply chain incident within a few months, following Codenotary's SBOM warnings in January. Security teams are being urged to pin LiteLLM to a verified version, rotate any credentials that may have been exposed, and audit transitive dependency trees for AI libraries across production environments. Read more

ByteDance Launches Dreamina Seedance 2.0 Inside CapCut, Bringing Studio-Grade Video AI to Mass Market

ByteDance rolled out Dreamina Seedance 2.0 directly inside CapCut this week, embedding advanced generative video tools into one of the world's most widely used editing apps. Users can generate high-quality video clips from text prompts, reference images, or short clips, with Seedance 2.0 delivering significantly improved motion consistency, physics simulation, and lighting coherence over its predecessor. Rather than launching as a standalone product, ByteDance chose CapCut — which has hundreds of millions of monthly active users — as the launch surface, making professional-grade AI video generation a feature of everyday mobile editing. The move intensifies pressure on Runway, Luma, and Kling, and positions ByteDance as the leading alternative now that OpenAI has exited the standalone video generation market with Sora's shutdown. Read more

Google Launches Gemini 3.1 Flash Live: Sub-Second Multimodal Responses for Real-Time App Developers

Google released Gemini 3.1 Flash Live this week, a new API tier optimized for real-time, low-latency interactions across voice, video, and text simultaneously. Designed for developers building live AI applications, Flash Live cuts time-to-first-token to under one second and maintains sub-200ms per-token latency at scale — making it viable for real-time voice agents, live visual assistants, and interactive coding tools. The release expands the multimodal speed Google demonstrated on the Galaxy S26 into the general API, giving all developers access without a Samsung device. Flash Live is available immediately in the Gemini API and Google AI Studio, with Vertex AI support following in the coming weeks. Read more

Want AI Updates Delivered Weekly?

Subscribe to our AI newsletter for the latest developments and insights.

Subscribe to Newsletter

Related Posts

How good is DeepSeek OCR?

How good is DeepSeek OCR?

DeepSeek-OCR goes beyond basic text extraction by understanding entire page layouts. We tested it on complex 19-page reports to evaluate its structural accuracy.

February 4, 2026 Read More →
Best Open Model for Real Prompts

Best Open Model for Real Prompts

Having tested top AI models on real-world tasks, GPT-OSS-120B leads in technical performance, Qwen3 excels at research, while GPT-5 and DeepSeek shine in coding and analysis. See the full benchmark results.

October 18, 2025 Read More →
Can ChatGPT create an app for me?

Can ChatGPT create an app for me?

ChatGPT can draft technical plans in minutes but needs human review. Here's how developers actually use AI coding tools to build apps faster.

December 9, 2025 Read More →