AI Newsletter February 2026

Welcome to the Dixon AI Newsletter which summarises selected AI stories and releases you might have missed. This issue covers AI news from 24th January 2026 to 22nd February 2026.

The AI Macro Environment and Enterprise Insights

• Agentic AI Moves from Pilot to Production Across Enterprise: A CrewAI survey of 500 large-enterprise leaders found 100 per cent plan to expand agentic AI adoption in 2026, with 31 per cent of workflows already automated. Gartner projects 40 per cent of enterprise applications will embed agents by year-end, up from under 5 per cent in 2025. Yet McKinsey finds fewer than one in four organisations have successfully scaled agents with integration complexity, governance gaps, and security as the primary blockers. The limiting factor is no longer capability; it is operational readiness. https://www.businesswire.com/news/home/20260211693427/en/Agentic-AI-Reaches-Tipping-Point-100-of-Enterprises-Plan-to-Expand-Adoption-in-2026-New-CrewAI-Survey-Finds
  

• Moltbook: The Governance Warning Hidden Inside a Viral Spectacle: Launched on 28 January, Moltbook - a Reddit-style forum where AI agents post and interact without human direction - reached 1.6 million registered agents within weeks. Researchers quickly concluded it was largely pattern-matching rather than genuine autonomy. More significant were the security findings: an exposed production database, prompt injection vulnerabilities, and credential theft risks. For enterprises deploying multi-agent systems, Moltbook is a vivid illustration of what ungoverned agentic infrastructure looks like at scale. https://www.technologyreview.com/2026/02/06/1132448/moltbook-was-peak-ai-theater/
  

• Anthropic and the Department of War Clash Over AI Safeguards: The Pentagon is threatening to designate Anthropic a "supply chain risk" - a status normally reserved for foreign adversaries - after months of failed negotiations over Claude's terms of use. Anthropic refuses to permit unrestricted military deployment, drawing the line at mass surveillance and fully autonomous weapons. The dispute, triggered by Claude's reported use in the January raid that captured Nicolás Maduro, sets a precedent for how AI companies govern their models in state-level military contexts. https://www.axios.com/2026/02/16/anthropic-defense-department-relationship-hegseth
  

• "Something Big Is Happening" and the Debate It Sparked: HyperWrite founder Matt Shumer's 5,000-word essay, published on 9 February, accumulated over 80 million views within days, arguing AI has crossed a threshold where white-collar displacement is imminent. Critics responded quickly: Fortune called it irresponsible, and Columbia professor Vishal Misra argued LLMs are probabilistic inference engines completing patterns, not exercising judgment. The episode matters less for its content than for what it revealed: public anxiety about AI and work has reached a mainstream inflection point. https://fortune.com/2026/02/11/something-big-is-happening-ai-february-2020-moment-matt-shumer/ 
  

• UK AI Skills Boost: The government expanded its AI Skills Boost programme in late January, raising its ambition from 7.5 million to 10 million UK workers to be upskilled by 2030 — roughly a third of the workforce. The free platform, delivered through the AI Skills Hub developed by PwC UK with Innovate UK, hosts over 660 courses benchmarked against Skills England's new AI Foundation Skills standard, with learners earning a government-backed virtual badge on completion. Over one million courses had been completed since the programme's original June 2025 launch. The expansion drew criticism, however: all 14 badged courses come exclusively from US tech firms including Google, Microsoft, and IBM, prompting accusations that the initiative deepens UK dependency on American platforms - the opposite of the government's stated ambition to make Britain an "AI maker, not an AI taker." https://www.gov.uk/government/news/free-ai-training-for-all-as-government-and-industry-programme-expands-to-provide-10-million-workers-with-key-ai-skills-by-2030n
  

New and updated AI model and application releases

• OpenAI 
  

  • GPT-5.3-Codex: Launched 5 February, combining the coding performance of GPT-5.2-Codex with the broader reasoning of GPT-5.2 in one model - around 25% faster. The first OpenAI release treated as "High capability" under its Preparedness Framework for cybersecurity, with a dedicated Trusted Access for Cyber programme for defenders. https://openai.com/index/introducing-gpt-5-3-codex/
    

  • GPT-5.3-Codex-Spark (research preview): A smaller, real-time version of GPT-5.3-Codex optimised for speed, delivering over 1,000 tokens per second. Available to Pro users in the Codex app, CLI, and IDE extension in partnership with Cerebras. https://developers.openai.com/codex/changelog/
    

  • OpenClaw acquisition: Peter Steinberger, the Austrian developer behind the viral open-source AI agent OpenClaw, joined OpenAI on 14 February to lead its next generation of personal agents. Sam Altman confirmed the hire on X, saying the work would "quickly become core to our product offerings." OpenClaw, which automates tasks across messaging platforms, email, and calendars using an LLM of the user's choice, will continue under an independent open-source foundation sponsored by OpenAI. https://techcrunch.com/2026/02/15/openclaw-creator-peter-steinberger-joins-openai/
    

  • Legacy model retirements: GPT-4o, GPT-4.1, GPT-4.1 mini, o4-mini, GPT-5 Instant, and GPT-5 Thinking were retired from ChatGPT on 13 February. No API changes at this time. https://help.openai.com/en/articles/6825453-chatgpt-release-notes
    

• Google
  

  • Gemini 3.1 Pro: Released 19 February, this upgraded core model scores 77.1% on ARC-AGI-2 — more than double its predecessor — and tops the APEX-Agents leaderboard for real-world knowledge work. Rolling out to AI Pro and Ultra subscribers, and available to developers via AI Studio, Vertex AI, and Google Antigravity. https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/
    

  • Lyria 3: Also released 19 February, Google's latest music generation model supports multiple genres and vocal languages, with imperceptible AI watermarking built in. Accessible via the Gemini app's Tools menu. https://gemini.google/release-notes/
    

• Microsoft
  

  • GPT-5.2 in Microsoft 365 Copilot: From late January, GPT-5.2 became available in the Copilot Chat model selector across all platforms, offering a Quick Response mode for fast answers and a Think Deeper mode for complex reasoning, grounded in work context via Work IQ. https://techcommunity.microsoft.com/blog/microsoft365copilotblog/what%E2%80%99s-new-in-microsoft-365-copilot--january-2026/4488916
    

  • Agent mode in Word, Excel, and PowerPoint: Agent mode - which actively edits and reasons through documents rather than simply responding to prompts - rolled out broadly in January and February across web and desktop for licensed users.https://techcommunity.microsoft.com/blog/microsoft365copilotblog/what%E2%80%99s-new-in-microsoft-365-copilot--january-2026/4488916
    

  • xAI Grok 4.1 Fast in Copilot Studio: From 19 February, US-based organisations can build with Grok 4.1 Fast in Copilot Studio, adding a fast-reasoning text model suited to large-context and complex tool-use workflows. Customer data is not retained or used to train xAI models. https://www.microsoft.com/en-us/microsoft-copilot/blog/copilot-studio/more-choice-more-flexibility-xai-grok-4-1-fast-now-available-in-microsoft-copilot-studio/

    
    

• Anthropic
  

  • Claude Cowork (expanded): Launched in research preview on 12 January and expanded to all Pro subscribers on 16 January, Cowork brings Claude Code's agentic capabilities to broader knowledge work — file organisation, data analysis, and multi-document project coordination — with eleven open-source role-specific plugins. https://www.anthropic.com/webinars/future-of-ai-at-work-introducing-cowork
    

  • Claude Opus 4.6: Released 5 February, Opus 4.6 improves on its predecessor across agentic coding, long-context reasoning, financial analysis, and professional document creation. It introduces a 1M token context window in beta, 128K output tokens, and takes the top spot on the Finance Agent and Terminal-Bench 2.0 benchmarks. https://www.anthropic.com/news/claude-opus-4-6
    

  • Claude in PowerPoint and Claude in Excel (updated): Released alongside Opus 4.6, Claude is now available as an add-in for PowerPoint, while the Excel integration was updated to use Opus 4.6 and now supports native operations including pivot table editing and conditional formatting. https://docs.anthropic.com/en/release-notes/claude-apps
    

  • Claude Sonnet 4.6: Released 17 February, priced identically to Sonnet 4.5. Delivers improved agentic search performance while consuming fewer tokens, with extended thinking and a 1M token context window in beta. Now the default model on claude.ai. https://www.anthropic.com/news/claude-sonnet-4-6
    

  • Claude Code Security: Introduced in February, this tool reviews codebases to identify vulnerabilities, extending Claude Code beyond generation and editing into active security analysis. https://www.anthropic.com/news/gov-UK-partnership

• Notable Other Model and Application Releases
  

  • SpaceX acquires xAI: SpaceX completed its acquisition of xAI in an all-stock deal valuing the combined entity at $1.25 trillion — the largest private merger on record. Musk cited orbital data centres as the primary rationale, arguing terrestrial infrastructure cannot meet AI's power demands at scale. The deal brings X under the SpaceX umbrella ahead of a planned IPO in mid-2026. https://techcrunch.com/2026/02/02/elon-musk-spacex-acquires-xai-data-centers-space-merger/
    

  • Grok Imagine 1.0: xAI released the Grok Imagine API, adding 10-second, 720p video generation with natively generated audio. The company reported 1.245 billion videos generated in January alone, and claimed top-of-table performance on speed and cost versus Sora and Veo on the Artificial Analysis benchmark. https://x.ai/news/grok-imagine-api
    

  • Grok 4.20 Beta: xAI launched a public beta of Grok 4.20, introducing a "rapid learning" architecture that updates the model weekly based on real-world feedback — a first for the Grok series. It adds a four-agent parallel collaboration system, medical document analysis via photo upload, and improved engineering reasoning. Available to SuperGrok and Premium+ subscribers; API access is listed as coming soon. https://x.ai/news
    

  • Seedance 2.0 (ByteDance): Released 10 February, a unified multimodal video generation model producing text-to-video and image-to-video clips with natively synchronised audio in a single generation pass. Went viral for celebrity deepfakes within days of launch, drawing cease-and-desist letters from Disney and Paramount; ByteDance subsequently tightened verification requirements. https://techcrunch.com/2026/02/15/hollywood-isnt-happy-about-the-new-seedance-2-0-video-generator/
    

  • Chinese model push: Several Chinese labs released flagship updates around the Lunar New Year holiday. ByteDance's Doubao 2.0 claims reasoning performance on par with GPT-5.2 and Gemini 3 Pro; Alibaba's Qwen 3.5 added agentic and multimodal capabilities at 60% lower cost than its predecessor, and the Qwen family has now overtaken Meta's Llama in cumulative Hugging Face downloads. Moonshot AI's Kimi K2.5 — a trillion-parameter MoE capable of coordinating up to 100 sub-agents in parallel — benchmarks close to Claude Opus 4.5 at roughly one-seventh the price. And Zhipu AI's GLM-5, a 744B MoE trained entirely on Huawei Ascend chips, signals a push for full independence from US semiconductor hardware. https://www.euronews.com/next/2026/02/17/these-are-chinas-new-ai-models-that-have-just-been-released-ahead-of-the-lunar-new-year
    

Recommended Reading, Listening and Book Releases

• For anybody:
  

  • TV Series: AI Confidential with Hannah Fry. A three part series on BBC iPlayer looking at some of the extraordinary human stories emerging from the world of AI. https://www.bbc.co.uk/programmes/m002q76b  
    

  • Book: The AI Transformation Playbook by Rob Dixon was published in December 2025 in both paperback and ebook formats. The AI Transformation Playbook

From Dixon AI

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

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

    • w/c 23rd Feb: What’s New in AI

    • w/c 2nd Mar: Claude Skills and Claude Cowork

    • w/c 9th Mar: Smarter Prompting

    • w/c 16th Mar: Excel with AI
      

• New Website: We launched a new website in December at www.dixonai.com and it’s continuing to evolve. Take a look and let us know what you think!