AI Newsletter March 2026

Welcome to the Dixon AI Newsletter which summarises selected AI stories and releases you might have missed. This issue covers AI news from 23rd February 2026 to 27th March 2026.

The AI Macro Environment and Enterprise Insights

• Anthropic vs the Pentagon (and the injunction that followed): The Trump administration designated Anthropic a supply-chain risk (a label typically reserved for foreign adversaries) after the company refused to allow Claude to be used for mass domestic surveillance or fully autonomous weapons. Anthropic sued the Department of Defense, citing First and Fifth Amendment violations. A federal judge in San Francisco granted a preliminary injunction blocking enforcement of the directive and halting the Pentagon's effort to designate Anthropic as a national security risk. The dispute has become the defining AI policy story of 2026 so far. See (Anthropic wins preliminary injunction in DOD fight as judge cites 'First Amendment retaliation')

• AI-driven job cuts accelerate: Two of the period's biggest workforce stories point in the same direction. Jack Dorsey's Block eliminated over 4,000 jobs (nearly half its workforce), citing AI as the sole driver and arguing that AI tools paired with smaller teams enable a fundamentally new way of working. Meanwhile, Meta is reportedly planning layoffs affecting 20% or more of its roughly 79,000 employees, driven by rising AI infrastructure costs and the expectation that AI-assisted workers will enable smaller teams. Both companies remain profitable and neither framed the cuts as a response to declining performance. See (theverge.com) and (theguardian.com)

• Anthropic labour market study shows early displacement signals: Anthropic published research introducing a new "observed exposure" metric combining theoretical LLM capability with real-world Claude usage data. The most exposed roles include computer programmers (75% task coverage), customer service representatives, and data entry workers (67%). Exposed workers tend to be older, female, more educated, and earn 47% more on average. Bureau of Labor Statistics projections show weaker job growth in exposed roles, and early data suggests hiring of younger workers has already slowed in those occupations. See (anthropic.com)

• US Supreme Court declines AI copyright case: The US Supreme Court declined to hear Stephen Thaler's case seeking copyright protection for AI-generated art, letting stand lower court rulings that AI-generated works cannot receive copyright. The decision sets an important precedent for how intellectual property law applies to AI-created content across creative industries. See (reuters.com)

• Wikipedia bans AI-generated articles: Wikipedia updated its English-language guidelines to ban editors from writing or rewriting articles using AI, citing AI-generated content's tendency to violate core content policies. Editors may still use large language models for basic copyediting suggestions, provided the AI does not introduce new content. The policy passed with overwhelming community support following months of cleanup efforts targeting AI-written articles. See (theverge.com)
  

New and updated AI model and application releases

• OpenAI
  

  • GPT-5.3 Instant: A faster, more conversational everyday model, improving on info-seeking, technical writing, and translation. Released 3rd March. GPT-5.3 Instant | OpenAI
    

  • GPT-5.4: OpenAI's most capable frontier model, available in Standard, Thinking, and Pro variants. Features a 1 million token context window, native computer-use capabilities, and 33% fewer factual errors than GPT-5.2. Released 5th March. Introducing GPT-5.4 | OpenAI
    

  • GPT-5.4 mini and nano: Smaller, cheaper models in the 5.4 family. GPT-5.4 mini is available to free and Go users via the Thinking feature; nano targets lightweight use cases. Released 17th March. GPT-5.4 mini and nano | OpenAI

    
    

  • ChatGPT for Excel: A new add-in powered by GPT-5.4 for building and analysing spreadsheet models using natural language. Includes new financial data integrations with Moody's, Dow Jones Factiva, MSCI, and others. ChatGPT for Google Sheets is also coming soon. ChatGPT for Excel | OpenAI
    

  • Codex Security: An application security agent (formerly Aardvark) for detecting and patching vulnerabilities. Reduced false positives by over 50% in beta. Now in research preview. Codex Security | OpenAI
    

  • Sora shutdown: OpenAI is discontinuing the standalone Sora video app and API. Video generation will continue within ChatGPT. Disney pulled out of its $1 billion Sora investment deal. Sora Shutdown
    

  • Agentic Commerce Protocol: Expanded shopping in ChatGPT with visual product discovery, side-by-side comparisons, and image-based search. Retailers including Walmart, Target, and Sephora have integrated. Product Discovery in ChatGPT | OpenAI

• Google

• Gemini 3.1 Flash-Lite: Google's fastest and most cost-efficient Gemini 3 model, released 3rd March. Gemini 3.1 Flash-Lite
  

• Gemini 3.1 Flash Live: A multilingual audio model powering more natural voice conversations across Google products, now expanded globally. Gemini 3.1 Flash Live
  

• Nano Banana 2 (Gemini 3.1 Flash Image): Advanced image generation combining the capabilities of Nano Banana Pro with Flash speed. Supports subject consistency for up to five characters and production-ready resolutions up to 4K. Nano Banana 2
  

• Lyria 3 Pro: An advanced AI music generation model creating tracks up to 3 minutes long with structural awareness (intros, verses, choruses). All outputs watermarked with SynthID. Lyria 3 Pro
  

• NotebookLM Cinematic Video Overviews: Powered by Gemini 3, Nano Banana Pro, and Veo 3, generating fluid animated video summaries from user sources. NotebookLM Video Overviews

• Microsoft

• GPT-5.4 Thinking in Microsoft 365 Copilot: Microsoft 365 Copilot now runs on OpenAI's GPT-5.4 Thinking model, available from 6th March. GPT-5.4 Thinking in M365 Copilot | Microsoft Community Hub
  

• Agentic Copilot in Excel, Word, and PowerPoint: Copilot can now take multi-step actions directly in files across all three apps, turning drafts into review-ready documents, trusted models, and on-brand presentations through natural conversation. Agentic Copilot | Microsoft Community Hub
  

• Work IQ: A new intelligence layer that personalises Microsoft 365 Copilot to individual users and their organisations, drawing on context from emails, meetings, and documents. Work IQ | Microsoft Community Hub
  

• Copilot Tasks: Shifts Copilot from answering questions to completing real-world actions, including recurring tasks, document generation, and logistics. In research preview and a waitlist. Copilot Tasks | Microsoft

• Anthropic

• Claude

  • 1M context window now generally available at standard pricing: The 1 million token context window is now generally available for Claude Opus 4.6 and Sonnet 4.6 with no long-context premium. Opus 4.6 at $5/$25 per million tokens; Sonnet 4.6 at $3/$15. Media limits expand to 600 images or PDF pages. 1M Context GA \ Anthropic
    

  • Inline visualisations: Claude can now generate interactive charts, diagrams, and visualisations directly within chat conversations (not as separate artifacts). Claude Builds Visuals \ Anthropic
    

  • Connectors and Memory on the free tier: Advanced features including remembering user preferences and connecting to external data sources are now available to free users. Memory on Free Plan \ Anthropic
    

  • Claude Skills for Excel and Slides: Advanced, specialised action sets for complex data manipulation in Excel and presentation design in PowerPoint, going beyond basic chat interaction. Claude Apps Release Notes \ Anthropic
    

• Claude Cowork

  • Computer use: Claude can now point, click, scroll, open files, use a browser, and run dev tools directly on macOS, with no setup required. Available for Pro and Max subscribers. Computer Use \ Anthropic
    

  • Dispatch: A routing system for assigning tasks from a phone and managing multi-agent “teams” within Cowork. Pairs with computer use so Claude can work on your behalf while you are away. Dispatch \ Anthropic
    

  • Cowork Projects: A structured way to organise long-term agent tasks and shared resources within the Cowork environment. Cowork Updates \ Anthropic
    

  • Cowork Plugins: Expansion of the Cowork ecosystem to allow third-party tool connections. Cowork Updates \ Anthropic
    

  • Scheduled Tasks: The ability to set Claude to perform recurring or future-dated actions automatically. Cowork Updates \ Anthropic

• Claude Code

  • Auto Mode: A new permissions setting allowing Claude to handle tool-call decisions autonomously, with a classifier blocking destructive operations. Available for Team plan users. Auto Mode \ Anthropic
    

  • Code Review: A multi-agent pull request review system, checking for bugs in parallel, filtering false positives, and ranking by severity. Reviews average 20 minutes. In research preview. Code Review \ Anthropic
    

  • Remote Control: Features allowing for more autonomous agentic coding across remote environments, extending Claude Code beyond the local terminal. Claude Code Updates \ Anthropic

• Notable Other Model Releases

• NVIDIA Nemotron 3 Super: A 120 billion parameter open model with only 12 billion active parameters at inference, delivering up to 5x higher throughput for agentic AI workloads. Features a million-token context window. Announced at GTC and available via Perplexity among other places. Nemotron 3 Super
  

• Alibaba Qwen 3.5 Small Series: Four dense open-source models (0.8B to 9B parameters), all natively multimodal. The 9B model scores 81.7% on GPQA Diamond, outperforming models 13x its size. Qwen 3.5 Small
  

• Mistral Voxtral TTS: Mistral's first text-to-speech model (4B parameters), supporting nine languages with emotionally expressive speech. Achieves 70ms model latency and outperforms ElevenLabs Flash v2.5 on naturalness. Voxtral TTS
  

• Luma Labs Uni-1: A new model that integrates reasoning and image generation in a single unified architecture, rather than handling them sequentially. https://lumalabs.ai/uni-1 

• Notable Application Releases

• OpenClaw on AWS Lightsail: AWS launched pre-configured OpenClaw instances on Lightsail with Amazon Bedrock, making it easy to run private autonomous AI agents. OpenClaw on Lightsail
  

• Perplexity Computer for Enterprise: An AI system running multi-step workflows across 20 models and 400+ apps, including Slack integration, aimed at enterprise automation. Perplexity Computer
  

• Adobe AI Assistant in Photoshop: A public beta allowing users to edit photos by typing or speaking natural language descriptions, plus access to 25+ AI models including OpenAI and Runway. AI in Photoshop
  

• Canva Magic Layers: Converts flat images and AI-generated visuals into fully editable, multi-layered designs. Now in public beta across the US, UK, Canada, and Australia. Magic Layers
  

Recommended Reading, Listening and Book Releases

For anybody:

• Podcast: Lex Fridman #494 with Jensen Huang (22nd March). Over two hours with the NVIDIA CEO covering scaling laws, data centre infrastructure, China, and the now-infamous moment where Huang claimed AGI has been achieved (before partially walking it back). Lex Fridman Podcast
  

• Podcast: The Artificial Intelligence Show (weekly). Paul Roetzer and Mike Kaput's weekly breakdown has been consistently strong through this period, covering the Anthropic/Pentagon saga, AI-driven layoffs, and the GPT-5.4 release in accessible depth. Apple Podcasts
  

• Podcast: Hard Fork by the New York Times (weekly). Kevin Roose and Casey Newton continue to do a great job of making AI news digestible for a general audience.

For the more technically minded and developers:

• Podcast: Latent Space (ongoing). Deep technical conversations with engineers building production AI systems. Particularly good recent episodes on agentic workflows and open-source model ecosystems. latent.space
  

• Open-source project: Andrej Karpathy's Autoresearch (released 10th March). An open-source script that automates AI experimentation overnight. In one run, the agent completed 126 experiments and reduced validation loss meaningfully. Worth exploring for anyone interested in how AI is accelerating AI research itself. GitHub

For leaders:

• Book: More Human by Rasmus Hougaard and Jacqueline Carter. Argues that AI adoption, done thoughtfully, can make leadership more human rather than less. Grounded in research and practical strategies for senior leaders.
  

From Dixon AI

• AI Drop In: one-hour, live sessions held twice a week. 

  • Join here: https://www.dixonai.com/ai-drop-in 

    • w/c 30rd Mar: What’s New in AI

    • w/c 7th Apr: Finding AI Use Cases
      

• AI Courses (full details on our website):

  • Our four-week Copilot Masterclass starts on 7th April

  • Our three-week AI Literacy Essentials course starts 13th April

  • A four-week Claude Masterclass starts on Wed 6th May
    

A word from Rob

If you've read through this newsletter, one thing should be clear: the pace isn't slowing down.

Knowledge workers are splitting into two groups. There are the AI-enabled, riding the current agentic AI revolution, compounding their advantage with every new release. And there's everyone else, watching the gap widen and wondering when things will settle down. They won't.

I can speak to this personally. I feel roughly 10x more productive than I did 12 months ago. But what's interesting is where the bottleneck has moved. I'm no longer execution bottlenecked. The constraint now is purpose and judgement, knowing what to build, why it matters, and what to say no to. I wrote about this shift in my recent working paper, From Knowledge to Purpose: The Shifting Bottleneck of Value Creation and Five Domains of Human Value as We Approach AGI, which explores how the locus of human value is changing as AI capability accelerates.

What makes this moment different is the rate of change. The distance between these two groups isn't growing steadily, it's accelerating. Every month that passes without action makes the next month's catch-up harder.

For leaders, this is no longer a technology decision. It's an existential one. The organisations investing in AI capability building now are pulling away from those still waiting for the "right time." That time was months ago. The train has left the station, don't miss the next one.

If you're ready to move, our AI Drop-In sessions, Copilot Masterclass, and AI Literacy Essentials course are all designed to close exactly this gap. The details are above, come and join us. Or reach out to me directly and we can have a conversation about helping your organisation catch this wave.

Rob