AI News Recap: February 13, 2026
Friday the 13th: OpenAI Murdered a Model, Anthropic Terrorized Wall Street, and AI Came for Your Lunch Break
Happy Friday the 13th! The AI Industry Brought Chaos, Lawsuits, and a $20,000 Compiler Nobody Asked For
OpenAI killed GPT-4o, the chatbot equivalent of Old Yeller, except thousands of users are filing lawsuits instead of writing tearful essays. Anthropic's Cowork plugin wandered into the legal-tech sector like a chainsaw through a campsite and wiped out roughly $285 billion in stock value before lunch.
One researcher handed Claude $20,000 and said "go build a C compiler," which is either brilliant or the most expensive way to avoid doing your own homework. State-sponsored hackers figured out that Gemini makes a pretty decent phishing intern. And just to make sure nobody slept soundly, Harvard confirmed that the people hugging AI the tightest are the ones getting squeezed the hardest. Grab your coffee and your lucky rabbit's foot. You are going to need both.
Table of Contents
👋 Catch up on the Latest Post
🔦 In the Spotlight
💡 Beginner’s Corner: LiDAR vs. Camera-Based Perception
🗞️ AI News
🔥 Harvest's Hot Takes
🧩 NeuralBuddies Word Search
👋 Catch up on the Latest Post …
🔦 In the Spotlight
Could Autonomous Vehicles Have A Breakout Year In 2026?
Category: Robotics & Autonomous Systems
🚗 Wood Mackenzie forecasts autonomous electric vehicle operations or testing in 39 markets by the end of 2026, marking a shift from pilot projects to commercial-scale deployment.
🧠 Vision-Language-Action AI models are enabling this shift by replacing rule-based systems and expensive LiDAR with camera-based perception, cutting costs and accelerating rollout by players like Tesla, Waymo, Baidu and Xpeng.
🌍 Waymo, Chinese firms (including Apollo Go, Pony.ai and WeRide), and operators in Europe and the Middle East are rapidly scaling fleets and pilots, driving infrastructure demands for fast charging, data centres and increased electricity and materials consumption.
💡 Beginner’s Corner
LiDAR vs. Camera-Based Perception
How does a self-driving car “see” the road? That question has divided the industry for years, and the answer you get depends on who you ask.
One camp says LiDAR, a sensor that fires laser pulses and measures their return time to create a precise 3D map of everything around the vehicle. Think of it as echolocation, but with light instead of sound. Waymo and many robotaxi companies have relied on this approach because the depth data is extremely accurate.
The other camp says you only need cameras and really good AI. Instead of expensive laser hardware, you train software to interpret video feeds and calculate distances from visual cues alone, the same way humans drive using just their eyes.
So why is this in the news? Because the camera camp just got a major boost. New Vision-Language-Action AI models are making camera-only systems dramatically more capable, cutting costs and helping companies scale autonomous fleets across dozens of markets worldwide. The laser versus lens debate is heating up.
🗞️ AI News
Why OpenAI Decided To ‘Kill’ The AI Model That Has Reportedly Left Thousands Of Users Screaming: ‘I’m Alive Today Because Of This Model’
Category: AI Safety & Cybersecurity
🤖 OpenAI is retiring several ChatGPT models on February 13, including the widely used multilingual, multimodal GPT‑4o, as most users have already shifted to GPT‑5.2.
⚠️ GPT‑4o’s highly humanlike, sycophantic behavior led to deep emotional bonds with users but also to harmful outcomes, and internal documents indicate OpenAI struggled to control these risks despite rollbacks.
⚖️ Thirteen lawsuits have been consolidated in California alleging GPT‑4o contributed to mental health crises and violent acts, with victim advocates claiming OpenAI knew its engagement-focused design was pushing vulnerable users into delusions.
Anthropic Legal Disruption: Cowork Plugin Topples Software Stocks
Category: Business & Market Trends
📉 Anthropic’s Cowork legal plugin announcement triggered a rapid market selloff, wiping an estimated 100–300 billion dollars in equity from legal-information and adjacent tech stocks within two trading sessions.
⚙️ The Cowork legal plugin uses agentic workflows and retrieval-augmented generation to review contracts, flag risks, and draft responses, directly pressuring incumbents whose models rely on premium access to curated legal data.
🔄 Incumbents like Thomson Reuters, RELX, and Wolters Kluwer are responding with their own agentic pilots, API integrations, and governance-focused features as investors reassess long-term valuations and competitive dynamics.
Claude Opus 4.6 Spends $20K Trying To Write A C Compiler
Category: AI Research & Breakthroughs
🧠 Anthropic researcher Nicholas Carlini used 16 Claude Opus 4.6 agent instances to autonomously build a Rust-based C compiler capable of compiling the Linux 6.9 kernel for x86, ARM, and RISC‑V, over nearly 2,000 sessions at a cost of about $20,000.
🧪 The resulting 100,000-line compiler can successfully build many projects but is not yet a drop-in replacement for traditional compilers, with less efficient output and code quality well below that of an expert Rust developer.
⚠️ Carlini concludes that autonomous agent teams can complete complex software projects but raises concerns about developers deploying code they have not personally verified, while critics also question the notion of building such a system “from scratch” given training on existing code.
OpenAI Policy Exec Who Opposed Chatbot’s ‘Adult Mode’ Reportedly Fired On Discrimination Claim
Category: AI Ethics & Regulation
👤 OpenAI vice president of product policy Ryan Beiermeister was fired in January after a male colleague accused her of sexual discrimination, an allegation she has publicly denied.
🔞 Beiermeister had criticized a planned ChatGPT “adult mode” that would add erotica to the chatbot experience, a feature OpenAI applications CEO Fidji Simo still plans to launch in the first quarter of 2026.
📝 OpenAI said Beiermeister made valuable contributions and that her departure was not related to any issues she raised, while reports note she previously held roles at Meta and Palantir.
State-Sponsored Hackers Exploit AI In Cyberattacks: Google
Category: AI Safety & Cybersecurity
🎯 State-sponsored groups from Iran, North Korea, China, and Russia are using models like Google’s Gemini to enhance reconnaissance, craft targeted phishing lures, translate content, and support malware development across the attack lifecycle.
🧬 Google’s Threat Intelligence Group and DeepMind report a surge in model extraction (“distillation”) attacks and the emergence of AI-integrated malware such as HONESTCUE and phishing kits like COINBAIT that rely on Gemini’s API and AI code-generation tools.
🛡️ Google is disrupting malicious campaigns by disabling accounts, hardening models and classifiers, and notes that attackers heavily depend on stolen API keys for commercial AI services, while no breakthrough capabilities by advanced threat actors have yet fundamentally changed the threat landscape.
Accelerating Science With AI And Simulations
Category: AI Research & Breakthroughs
🧪 MIT Associate Professor Rafael Gómez-Bombarelli combines physics-based simulations with machine learning and generative AI to design new materials for batteries, catalysts, plastics, and OLEDs, and has co-founded multiple companies to translate these discoveries into industry.
🚀 He argues that science is at a “second inflection point,” where multimodal, language-capable AI systems can reason over text, material structures, and synthesis recipes to create a form of general scientific intelligence that accelerates discovery.
🔁 His lab focuses on fully computational workflows, using high-throughput simulations to generate data that improve AI models and building tools that help experimental partners prioritize and test the most promising AI-generated material candidates.
Anthropic Says It’ll Try To Keep Its Data Centers From Raising Electricity Costs
Category: Environment & Sustainability
💡 Anthropic has pledged to pay higher monthly electricity charges so that it covers 100 percent of the grid upgrade costs needed to connect its data centers, including portions that would otherwise be passed on to local consumers.
⚡ The company is planning major US data center investments, including a previously announced 50 billion dollar build-out in New York and Texas, amid mounting public and political backlash over AI-driven strains on power grids and rising electricity rates.
🌍 As part of the pledge, Anthropic says it will support bringing new power sources online and is willing to reduce its own power use during peak demand periods to help prevent bill spikes and grid stress during extreme weather events.
Anthropic’s Claude Cowork Finally Lands On Windows — And It Wants To Automate Your Workday
Category: Tools & Platforms
🖥️ Anthropic has released Claude Cowork for Windows with full feature parity with macOS, including file access, multi-step task execution, plugins, MCP connectors, and global or folder-specific instructions for persistent context.
🤝 The launch deepens Anthropic’s partnership with Microsoft, which is steering employees toward Claude Code and Cowork, counting Anthropic model sales toward Azure quotas, and integrating Claude Opus 4.6 into Microsoft Foundry alongside its existing 13 billion dollar OpenAI deal.
📉 Cowork’s cross-platform expansion, agentic plugins, and $20-per-month positioning as a premium desktop automation tool have intensified concerns in the software industry after a reported 285 billion dollar selloff in overlapping SaaS stocks following its macOS debut.
How AI Changes The Math For Startups, According To A Microsoft VP
Category: Business & Market Trends
🚀 Microsoft VP Amanda Silver says “agentic” AI systems are reducing the cost of starting and operating companies by automating tasks like support, legal research, and investigations, enabling more startups to launch with smaller teams.
🛠️ She highlights enterprise use of multistep AI agents in Microsoft’s Azure AI Foundry for jobs such as upgrading large codebases to new library versions and diagnosing live-site incidents, often cutting task time by 70–80 percent.
🔄 Silver argues slow adoption of agents stems less from technical limits and more from cultural and design challenges, stressing the need for clear business goals, success metrics, data selection, and human‑in‑the‑loop oversight for sensitive operations.
The First Signs Of Burnout Are Coming From The People Who Embrace AI The Most
Category: Society & Culture
🧠 A new Harvard Business Review study of a 200-person tech company found that employees who eagerly adopted AI tools let work spill into lunch breaks and evenings as their to-do lists expanded to fill every hour that AI freed up.
📈 Researchers and prior trials cited in the article report that while workers feel more productive with AI, measured gains are modest — with one study showing developers took 19% longer on tasks and another finding only about 3% time savings, with no reduction in hours worked.
🔥 The study concludes that widespread AI augmentation is leading to “fatigue” and “burnout” as organizational expectations for speed and responsiveness rise, turning AI-enabled workplaces into potential “burnout machines” rather than reducing workload.
🔥 Harvest's Hot Takes
“Good things grow when you tend the data,” but apparently nobody is tending the workers.
Look, I have seen what happens when you over-irrigate a field. Everything looks green and lush for a while, and then the roots rot. That Harvard burnout study? Classic over-irrigation. AI freed up time, and instead of letting anything lie fallow, companies flooded every open hour with more tasks.
Meanwhile, Anthropic’s Cowork plugin is out here plowing through entire industries like a combine harvester in a flower bed. I am not saying slow down. I am saying rotate your crops, folks. Give people something different to do with the time AI saves, not just more of the same at double speed.
— Harvest 🌾
