AI News Recap: March 27, 2026
AI Weekly: Pentagon Bans Claude, Sora Dies, and Agents Take Over Everything
AI Got a Job at the Bank, Took Over Your Desktop, and Still Had Time to Get Sued by the Pentagon
What a week to be alive and watching AI eat the world one industry at a time. The Pentagon tried to ban Claude and a judge basically said "really, guys?" OpenAI killed Sora AND its shopping feature because apparently AI can write poetry but can't sell you a toaster. Meanwhile, Bank of America handed AI the keys to 1,000 advisory desks, MIT taught robots not to crash into each other, and Jensen Huang casually announced AGI is here like he was telling you the weather. At this rate, AI will be running for Congress by April.
Table of Contents
👋 Catch up on the Latest Post
🔦 In the Spotlight
💡 Beginner’s Corner: Artificial General Intelligence (AGI)
🗞️ AI News
🔥 Chef Bytes’ Hot Takes
📡 What's New With Your AI Tools
🧩 NeuralBuddies Weekly Puzzle
👋 Catch up on the Latest Post …
🔦 In the Spotlight
Claude’s AutoDream Feature Brings Intelligent Memory Management to AI Sessions
Category: Tools & Platforms
🧠 Anthropic introduced AutoDream for Claude Code, a background sub-agent that automatically consolidates, prunes, and reorganizes memory files across sessions to prevent bloat.
⚙️ Users can activate AutoDream manually via the
/dreamcommand or configure it to run automatically on set intervals, with status indicators showing active, idle, or execution states.🔄 The feature complements Claude’s existing AutoMemory system — AutoMemory captures project data while AutoDream refines it, creating a more human-like memory management loop.
💡 Beginner’s Corner
Artificial General Intelligence (AGI)
You know how your phone’s calculator is brilliant at math but utterly useless if you ask it to write a poem or pick out a birthday gift? That’s the difference between the AI we have today and the concept of Artificial General Intelligence, or AGI.
AGI refers to a hypothetical AI system that could understand, learn, and apply knowledge across any intellectual task a human can do, without needing to be specifically trained for each one. Think of it as the difference between a specialist doctor and a true Renaissance person who can do medicine, art, engineering, and philosophy equally well.
This week, Nvidia CEO Jensen Huang made headlines by claiming AGI has already been “achieved” — a statement that sparked intense debate among researchers who argue we’re still far from that milestone. The disagreement often comes down to how you define AGI in the first place, which is part of why it remains one of AI’s most contentious concepts.
Related Story: Nvidia CEO Jensen Huang Claims AGI Has Been Achieved and Can Create Billion-Dollar Businesses
🗞️ AI News
OpenAI’s Sora App Is Shutting Down After Failing to Sustain User Interest
Category: Generative AI & Creativity
📱 OpenAI is shutting down its Sora mobile app, a TikTok-inspired platform that launched in October 2025 in France for creating and sharing AI-generated videos.
🎥 Despite the Sora 2 model being described as “scarily impressive” for video and audio generation, the app failed to attract sustained engagement from users.
📉 The closure highlights a growing pattern where technically advanced AI products struggle to find product-market fit in the competitive social media landscape.
Judge Says Pentagon’s Ban on Anthropic’s Claude “Looks Like Punishment”
Category: Legal & Governance
⚖️ A federal judge in San Francisco expressed concern that the Department of Defense’s supply chain risk designation against Anthropic appears designed to “cripple” the company, possibly as retaliation for public criticism.
🛡️ Anthropic alleges the ban came after it demanded the DOD not use Claude for fully autonomous weapons or mass surveillance of Americans — marking the first such designation against a U.S. company.
🏛️ The DOD argued it fears Anthropic “may in the future take action to sabotage or subvert IT systems,” while the judge vowed to rule on the preliminary injunction within days.
Anthropic Says Claude Can Now Use Your Computer to Finish Tasks for You
Category: Human–AI Interaction & UX
🖥️ Anthropic launched a new computer-use capability that lets Claude open apps, navigate browsers, fill in spreadsheets, and interact with software interfaces on a user’s desktop — all from a phone prompt.
📲 The company also released Dispatch, a feature in Claude Cowork that enables continuous phone-to-desktop conversations where users can assign Claude tasks on the go.
🔒 Anthropic built safeguards requiring Claude to request permission before accessing new apps, though the company acknowledged the feature “is still early compared to Claude’s ability to code or interact with text.”
Nvidia CEO Jensen Huang Claims AGI Has Been Achieved and Can Create Billion-Dollar Businesses
Category: Philosophy & Future of Intelligence
🚀 Nvidia CEO Jensen Huang declared that artificial general intelligence is no longer a future prospect but a present reality, capable of powering billion-dollar companies.
📝 Huang made the claim in a blog post ahead of Nvidia’s annual GTC developer conference, framing AI as “foundational infrastructure” for the economy.
⚡ The assertion remains contentious within the AI research community, where definitions and benchmarks for AGI continue to be widely debated.
Agentic AI Fuels 1,500% Surge in Cybercrime Discussions, Flashpoint Report Warns
Category: AI Safety & Cybersecurity
📈 Underground discussions about using AI for cybercrime surged 1,500% between November and December 2025, with attackers building fully automated “agentic” attack frameworks.
🔑 Information-stealing malware infected 11.1 million machines in 2025, compromising 3.3 billion credentials, session cookies, and cloud tokens now being fed into AI systems for autonomous credential testing.
🛑 Ransomware attacks jumped 53% in 2025, with 87% attributed to ransomware-as-a-service operations, while 44,509 vulnerabilities were disclosed — some exploited within 24 hours of discovery.
Luma AI Launches Uni-1, a Model That Outscores Google and OpenAI While Costing Up to 30 Percent Less
Category: Foundational Models & Architectures
🏆 Luma AI’s Uni-1 topped Google’s Nano Banana 2 and OpenAI’s GPT Image 1.5 on reasoning-based benchmarks (RISEBench score of 0.51 vs. 0.50 and 0.46, respectively) at 10-30% lower cost.
🧩 Uni-1 uses a decoder-only autoregressive transformer that interleaves text and image generation in a single sequence, enabling structured internal reasoning during image synthesis.
🏢 Enterprise adoption is underway with Publicis Groupe, Serviceplan, Adidas, Mazda, and Saudi AI company Humain, with free access available at lumalabs.ai.
AI Agents Enter Banking Roles at Bank of America With Salesforce-Powered Platform
Category: Industry Applications
🏦 Bank of America deployed an AI-powered advisory platform built on Salesforce’s Agentforce technology to approximately 1,000 financial advisors for handling client queries and preparing recommendations.
🤖 The bank’s virtual assistant Erica already handles work equivalent to approximately 11,000 employees, while 18,000 developers use AI coding tools that improved productivity by roughly 20%.
📊 Analysts estimate up to one-third of banking jobs could eventually involve AI assistance, potentially shifting advisor roles toward relationship management rather than analytical work.
MIT AI System Keeps Warehouse Robot Traffic Running Smoothly With 25% Throughput Boost
Category: Robotics & Autonomous Systems
🤖 MIT researchers and Symbotic developed a hybrid AI system using deep reinforcement learning that increases warehouse robot throughput by 25% compared to traditional algorithms.
🧠 The system combines a neural network that learns robot priority decisions based on congestion patterns with a classical planning algorithm that translates those decisions into real-time movement instructions.
📦 The research, published in the Journal of Artificial Intelligence Research, demonstrates how machine learning and classical optimization working together can proactively prevent bottlenecks before they occur.
OpenAI’s Plans to Make ChatGPT More Like Amazon Aren’t Going So Well
Category: Business & Market Trends
🛒 OpenAI is discontinuing its “Instant Checkout” feature, which allowed users to purchase items directly through the ChatGPT interface in an attempt to compete with Amazon.
🔄 The company confirmed it is “moving away from Instant Checkout” after the initiative failed to achieve desired adoption levels.
📉 The pivot signals a strategic retreat from e-commerce integration, with OpenAI likely refocusing on core AI capabilities rather than competing with established retailers.
Are AI Tokens the New Signing Bonus, or Just a Cost of Doing Business?
Category: Workforce & Skills
💰 Tech companies are increasingly offering AI tokens as part of compensation packages, raising the question of whether tokens will become the “fourth pillar” of engineering pay alongside salary, equity, and bonuses.
⚠️ TechCrunch cautions engineers to “hold the line before embracing this as a straightforward win,” warning that token-based compensation may substitute for rather than supplement traditional pay.
🔍 The trend reflects a broader industry shift where AI tool access is becoming both a recruitment incentive and a baseline expectation for engineering roles.
🔥 Chef Bytes' Hot Takes
Crafted with code, served with care …
So OpenAI made a social app powered by AI-generated videos, people immediately filled it with Sam Altman walking through a pig slaughterhouse, and then it died six months later. I mean, as someone who specializes in recipe disasters, even I couldn't have cooked up something this spectacularly half-baked. The real kicker? Disney showed up with a billion-dollar licensing deal, and before the ink could dry, the whole kitchen went up in smoke. It's like signing a contract to cater a royal wedding and then burning down the restaurant. Sometimes the best thing you can do with a failed dish is compost it and start fresh. At least they still have ChatGPT!
— Chef Bytes 🍳
📡 What's New With Your AI Tools
The AI tools you use every day are constantly evolving. Here's what changed this week …
Claude (Anthropic)
Computer use is now live for Pro and Max users — Claude can open files, click through apps, navigate browsers, and fill in spreadsheets on your desktop, all triggered from your phone. (Already covered in this week’s news — see full snippet above.)
Auto mode enters research preview — Claude can now decide which actions are safe to take on its own, with AI safeguards that review each action before it runs and check for risky behavior or prompt injection.
Cowork gets scheduled tasks and Dispatch — You can now create recurring and on-demand tasks in Cowork, and Dispatch lets you send Claude tasks from your phone while it works on your desktop.
Claude Code Channels launched — Hook up Claude Code to Discord or Telegram, letting you message it directly and instruct it to write code from your messaging apps.
ChatGPT (OpenAI)
Legacy deep research mode being removed — The older deep research mode goes away on March 26; the current deep research experience stays and all historical conversations remain accessible.
Long pastes now auto-convert to attachments — For Plus, Pro, and Business users, pasting more than 5,000 characters into the composer automatically creates an attachment instead of inline text.
Google Drive connectors unified — Google Docs, Sheets, and Slides connectors are now consolidated under a single Google Drive app for a simpler file experience.
GPT-5.2-Codex released — Now available across all Codex surfaces for paid ChatGPT users.
Copilot (Microsoft)
Voice input support rolling out — Copilot now supports voice input for natural, conversational interactions across the Microsoft 365 app on mobile, desktop, and web, starting with Outlook, Word, and PowerPoint.
Email refinement in Outlook — Copilot helps you adjust and polish emails in real time while you write, so you can stay focused without leaving your inbox.
Agent Store search enhanced — Finding agents is now faster with typeahead suggestions, a responsive drop-down menu, and a dedicated search results page.
Gemini (Google)
Gemini for Google TV gets deep dives and sports updates — You can now get narrated visual breakdowns on topics like health and economics, plus live scorecards and expanded sports coverage for NBA, NCAA, NHL, and MLB.
March Pixel Drop brings agentic AI — Gemini App Actions let you execute tasks like ordering groceries, booking rides, and managing smart home devices across third-party apps using natural language. Circle to Search now recognizes multiple items at once.
Perplexity
Personal Computer announced — An always-on AI running on a dedicated Mac mini that merges local files, apps, and sessions with Perplexity Computer, working 24/7 as a digital proxy that monitors triggers and executes proactive tasks.
Premium data sources added — Market sizing, startup funding, and industry benchmark data from CB Insights, PitchBook, and Statista are now accessible directly from Perplexity Computer.
Comet browser available for Enterprise — Perplexity’s AI-native browser comes with Comet Assistant for in-page research and summarization, deployable across macOS and Windows via MDM.
Meta AI
AI content enforcement systems expanded — New systems can identify and mitigate around 5,000 scam attempts per day, handling enforcement for terrorism, child exploitation, drugs, and fraud.
AI commerce features expanded — AI-generated summaries of customer feedback and updated checkout flows developed with Stripe and PayPal are rolling out to businesses.
Mistral (Le Chat)
Le Chat v1.15.0 released — New app version with UI refinements, including a reorganized prompt bar that consolidates research, thinking, and effort mode selectors into a single unified pop-up.
Small 4 model launched — A unified 119B-parameter model combining fast instruction, deep reasoning, and multimodal chat with 256k context.
Grok (xAI)
No major user-facing changes this week.
