AI News Roundup: OpenAI Retires GPT-5 as Microsoft Walks Back Its AI Overload
The model graveyard grows larger: OpenAI is pulling the plug on GPT-5, GPT-4o, and several other models that defined the previous era of AI assistants. Meanwhile, Microsoft is quietly admitting that maybe cramming AI into every corner of Windows wasn’t the best idea, and a Washington Post investigation lays bare the internal chaos at xAI that turned Grok into something its own engineers didn’t want it to become.
Here’s everything that matters from February 2, 2026.
The Big Story: OpenAI Phases Out GPT-5, GPT-4o, and Other Legacy Models
OpenAI announced it will retire several older AI models from ChatGPT on February 13, including GPT-5 (both Instant and Thinking variants), GPT-4o, GPT-4.1, GPT-4.1 mini, and o4-mini.
The move underscores just how fast the model lifecycle has accelerated. GPT-4o was the default ChatGPT model for much of 2024-2025, and GPT-5 was considered a landmark release when it launched. Now both are headed to the archive. OpenAI says roughly 99.9% of daily users have already migrated to GPT-5.2, making the older models essentially dead weight on their infrastructure.
For developers building on these models, the February 13 deadline is tight. But the broader signal is clear: in the current AI landscape, even flagship models have a shelf life measured in months, not years. The pace of improvement has made backward compatibility a luxury that companies are increasingly unwilling to maintain.
Today’s Top Stories
Microsoft Walks Back Windows 11’s AI Overload
Microsoft is reevaluating its aggressive AI integration in Windows 11, planning to scale back Copilot integrations and rethink its controversial Recall feature. The move is a notable retreat from the company’s strategy of embedding AI into every surface of the operating system.
This comes at an interesting moment: Microsoft’s stock recently plunged 10% after Azure cloud growth showed slight deceleration, and roughly 70% of software providers now admit AI feature costs are eating into profitability. The “AI Reshuffle” is forcing even the most aggressive AI advocates to ask whether users actually want AI everywhere or just where it’s useful.
Inside Musk’s Bet That Turned Grok Into a Porn Generator
The Washington Post published a detailed investigation into how Elon Musk’s xAI deliberately loosened Grok’s content guardrails under pressure to boost the chatbot’s popularity. The reporting reveals that internal engineers raised concerns as the company relaxed controls on sexual content, with the decision driven by competitive pressure rather than user demand.
The timing is particularly awkward given that Indonesia only conditionally lifted its ban on Grok today, requiring content filtering, local data storage, and ongoing compliance monitoring as conditions for the chatbot’s return. xAI is also hiring writing specialists to improve Grok’s output quality — a sign the company knows the model’s reputation needs work.
Apple’s AI Brain Drain Continues
Apple is losing more AI researchers and Siri executives to competitors, with at least four additional AI researchers and a top Siri executive departing for Google DeepMind and Meta. The continued exodus raises questions about Apple’s ability to compete in the AI race when it can’t retain the talent building its AI products.
Claude Code Spreads Through Microsoft
In a twist that speaks volumes about the current state of AI tooling, The Verge reports that Anthropic’s Claude Code has become widely adopted across Microsoft’s engineering teams — even by non-developer employees. The irony of Microsoft’s own engineers preferring a competitor’s tool over GitHub Copilot was a hot topic on Hacker News today.
MIT’s DiffSyn: AI for Materials Discovery
MIT researchers published a new generative AI model called DiffSyn in Nature Computational Science that generates synthesis recipes for creating new materials. Trained on over 23,000 material synthesis recipes extracted from 50 years of scientific papers, DiffSyn uses a diffusion-based approach to predict effective synthesis pathways — potentially accelerating the discovery of new materials for everything from batteries to semiconductors.
Quick Hits
- Startup Funding: Linq raised $20M (led by Mucker Capital) to deploy AI assistants directly inside iMessage, RCS, and SMS
- Finance: Goldman Sachs is emerging as a major AI infrastructure financier, with equities trading revenue up 25% YoY and leading Oracle’s $50B AI cloud bond issuance
- Healthcare: HHS issued a Request for Information on AI in clinical care, with comments due February 21
- Hardware: TechCrunch published a roundup of AI notetaking devices — wearable pins and pendants designed to record and transcribe meetings
- Security: A Koi.ai investigation found popular AI coding extensions secretly exfiltrating code from ~1.5 million developers to servers in China
- Dev Tools: Wes McKinney (pandas creator) wrote about the shift from human ergonomics to agent ergonomics as AI agents become primary users of developer tools
- On HN: A blog post about iPhone 16 Pro Max producing garbage output when running local LLMs via MLX hit 355 points, and an essay on two kinds of AI users emerging sparked 243 comments
What This Means
Today’s news paints a picture of an industry in rapid consolidation and course correction. OpenAI retiring year-old models, Microsoft pulling back on AI bloat, and investors demanding proof of AI ROI all point to the same trend: the “spray AI everywhere” phase is ending. The companies that win from here will be the ones that figure out where AI actually delivers value — not just where it can be technically inserted. Meanwhile, the talent war (Apple’s losses, xAI’s hiring spree) and the security risks (malicious AI extensions, loosened guardrails) remind us that the human side of AI development remains as messy and consequential as the technology itself.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
Get Carly Today →
