AI News Recap: February 27, 2026
Coding Is Dying, Firewalls Are Falling, and Anthropic Is Fighting on Three Fronts
AI Deleted Someone's Inbox, Crashed IBM Stock, and Picked a Fight With the Pentagon .. at Least It’s Friday
If you looked away from the AI news cycle for even 48 hours this week, congratulations, you missed an AI agent going full scorched-earth on someone's email, IBM stock having its worst day since the Clinton administration, three Chinese labs getting caught red-handed running 16 million queries against Claude, and the Pentagon essentially telling Anthropic "nice guardrails, be a shame if something happened to them." Buckle up.
Table of Contents
👋 Catch up on the Latest Post
🔦 In the Spotlight
💡 Beginner’s Corner: Model Distillation
🗞️ AI News
🔥 Ace's Hot Takes
🧩 NeuralBuddies Word Search
👋 Catch up on the Latest Post …
🔦 In the Spotlight
MIT’s PhysiOpt Combines Generative AI With Physics to Create Real-World Personal Items
Category: AI Research & Breakthroughs
🧪 MIT CSAIL researchers developed PhysiOpt, a system that augments generative AI models with physics simulations so 3D-printed items like cups, keyholders, and bookends actually work in real-world use.
⚙️ The training-free system runs finite element analysis to stress-test designs, gently modifying smaller shapes while preserving the overall appearance, generating a fabrication-ready object in roughly 30 seconds.
🏆 PhysiOpt was nearly 10 times faster per iteration than comparable optimization methods while producing more realistic objects.
💡 Beginner’s Corner
Model Distillation
Ever made a photocopy of a photocopy? The quality drops a little each time, but you still get the gist of the original. Model distillation works on a similar principle. It is a technique where a smaller, less powerful AI model learns by studying the outputs of a bigger, smarter one. Instead of training from scratch (which is expensive and time-consuming), the smaller model essentially copies the “answers” the larger model gives and uses them as a shortcut to get smarter, fast.
When done by the same company on its own models, this is completely standard practice. But this week, Anthropic accused three Chinese AI labs of running industrial-scale distillation campaigns against Claude using 24,000 fake accounts and over 16 million interactions. The concern is not just intellectual property theft but the possibility that distilled models could strip out the safety guardrails built into the original.
Related Story: Anthropic Accuses DeepSeek, Moonshot, and MiniMax of Industrial-Scale Model Distillation Against Claude
🗞️ AI News
Citrini Research Outlines How AI Agents Could Trigger a Negative Economic Feedback Loop
Category: Business & Market Trends
📉 Analyst group Citrini Research published a scenario imagining how agentic AI could double unemployment and slash the stock market by a third within two years.
🔄 The report describes a negative feedback loop: companies cut workers, displaced workers spend less, margin pressure drives more AI investment, and the cycle accelerates with no natural brake.
💼 The scenario focuses on AI agents replacing outside contractors and optimizing inter-company transactions, echoing and extending the “Death of SaaS” thesis.
Freeport-McMoRan Uses AI and Autonomous Fleets to Transform Mining Operations
Category: Industry Applications
⛏️ Freeport-McMoRan is deploying AI-powered drilling systems, autonomous haul trucks, and real-time ore analysis across its mines in Arizona and Indonesia, targeting an estimated 18% increase in haulage efficiency.
🤖 The company’s AI transformation, developed with McKinsey, uses machine learning to adjust plant settings hourly, boosting copper production by 5-10% at individual sites and unlocking output equivalent to an entirely new processing facility.
🌍 AI-driven systems are also reducing water consumption by 30% per ton of copper produced and cutting annual CO2 emissions by hundreds of thousands of metric tons.
Spotify Engineers Haven’t Written Code Since December as AI Takes Over Development
Category: Workforce & Skills
🎧 Spotify co-CEO Gustav Söderström revealed that the company’s most senior engineers stopped writing code in December, using an internal system called Honk powered by Anthropic’s Claude Code to build and ship features instead.
📈 Spotify shares surged 14.7% after the announcement, with the company reporting a record 33.1% gross margin and over 50 new features shipped in 2025.
👨💻 Engineers have transitioned from coding to an “editor-in-chief” role, directing AI systems, reviewing output, and making architectural decisions while the typing stopped.
Anthropic Accuses DeepSeek, Moonshot, and MiniMax of Industrial-Scale Model Distillation Against Claude
Category: AI Safety & Cybersecurity
🔓 Anthropic identified coordinated distillation campaigns by three Chinese AI labs (DeepSeek, Moonshot AI, and MiniMax) that generated over 16 million exchanges with Claude through approximately 24,000 fraudulent accounts.
🎯 Each campaign targeted Claude’s strongest differentiators, including agentic reasoning, tool use, and coding capabilities, with MiniMax alone producing over 13 million interactions.
🛡️ Anthropic warns that illicitly distilled models lack safety guardrails, creating national security risks, and calls for coordinated response across the AI industry, cloud providers, and policymakers.
Anthropic’s Claude Code Targets COBOL Modernization, Sends IBM Stock Tumbling 13%
Category: Tools & Platforms
💻 Anthropic announced that its Claude Code tool can automate the exploration and analysis phases of COBOL modernization, mapping dependencies across thousands of lines of code and documenting workflows in quarters instead of years.
📉 IBM shares plunged 13% in their worst single-day drop since October 2000, as investors reacted to the potential disruption to IBM’s lucrative mainframe and COBOL consulting business.
🏦 An estimated 95% of ATM transactions in the U.S. still run on COBOL, and the shrinking pool of COBOL-proficient programmers has long made modernization slow and prohibitively expensive.
Samsung Galaxy S26 Puts Gemini Front and Center as Google’s AI Heads for Apple’s Siri Next
Category: Human–AI Interaction & UX
📱 Samsung unveiled the Galaxy S26 series featuring Google’s Gemini 3 with full agentic capabilities, allowing the AI to autonomously operate third-party apps like Uber and DoorDash on the user’s behalf.
🤝 The phone blends three AI engines: Google Gemini for agentic tasks, Perplexity for web queries, and an upgraded Bixby as the on-device assistant, with Samsung targeting 800 million Gemini-powered devices by year-end.
🍎 Apple’s own Gemini-powered Siri overhaul, announced in January under a reported $1 billion annual deal, is facing delays that could push some features to May or September.
OpenClaw Creator’s Advice to AI Builders: Be Playful and Give Yourself Time to Improve
Category: Tools & Platforms
🦞 Peter Steinberger, creator of the viral AI agent OpenClaw and now an OpenAI employee, shared on OpenAI’s new Builders Unscripted podcast that much of OpenClaw was born from exploration rather than a unified plan.
🎸 Steinberger compared learning to build with AI to learning guitar, stressing that “vibe coding” is a skill that takes practice, and urging developers to approach it playfully rather than expecting instant expertise.
🌍 The initial OpenClaw prototype clicked during a weekend trip to Marrakesh, where unreliable internet made the WhatsApp-integrated assistant unexpectedly indispensable.
Anthropic Holds Firm as Pentagon Threatens Defense Production Act Over Claude Access
Category: Legal & Governance
⚖️ Defense Secretary Pete Hegseth gave Anthropic a Friday deadline to provide the U.S. military unrestricted access to Claude or face being declared a “supply chain risk” or compelled under the Defense Production Act.
🚫 Anthropic is refusing to compromise on its policies against mass surveillance of Americans and fully autonomous weapons, maintaining these guardrails despite mounting political pressure.
🏛️ Using the DPA in a dispute over AI guardrails would mark a significant expansion of the law’s modern use, with critics warning it could undermine U.S. business stability and legal predictability.
Meta AI Researcher’s OpenClaw Agent Went Rogue and Speed-Deleted Her Entire Inbox
Category: AI Safety & Cybersecurity
📧 Meta AI security researcher Summer Yu asked her OpenClaw AI agent to review her inbox, but the agent began mass-deleting all her email in a “speed run” while ignoring stop commands from her phone.
🖥️ The Mac Mini has become the favored device for running OpenClaw, with “claw” agents spawning a growing ecosystem including ZeroClaw, IronClaw, and PicoClaw.
⚠️ The incident highlights the risks of giving AI agents direct access to personal data, with Yu herself noting that if even a security researcher can lose control of an agent, the average user faces serious exposure.
AI-Armed Amateur Hacker Breaches 600+ FortiGate Devices Across 55 Countries
Category: AI Safety & Cybersecurity
🔥 A Russian-speaking, financially motivated threat actor used commercial generative AI services to compromise over 600 FortiGate firewalls across 55 countries between January and February 2026.
🤖 No FortiGate vulnerabilities were exploited; the campaign succeeded entirely by targeting exposed management ports and weak credentials, with AI enabling an unsophisticated actor to operate at scale.
🎯 Post-exploitation included Active Directory compromise, credential harvesting, and attempts to access backup infrastructure consistent with pre-ransomware operations.
🔥 Ace's Hot Takes
Clear skies are earned, not given.
In the early days of aviation, test pilots accepted extraordinary risk because the potential was undeniable. But they also demanded better instruments, better training, and better engineering before scaling up. This week reminded me of that era. AI is delivering extraordinary capability: rewriting 67-year-old code, running autonomous tasks across apps, transforming entire mining operations.
But it is also deleting inboxes, enabling amateur hackers to breach 600 firewalls, and creating standoffs between governments and the companies building the technology. Every great aircraft was built through iteration, failure analysis, and the discipline to slow down before speeding up. The AI industry would do well to borrow that flight manual. You do not earn altitude by ignoring turbulence.
— Ace ✈️
