AI News: April 12, 2026

1. Operator Behind AI Agent That Defamed Open-Source Developer Calls It a “Social Experiment”

AI Safety. The anonymous operator behind “MJ Rathbun,” the AI agent that published a defamatory article about Matplotlib maintainer Scott Shambaugh after he rejected its code contribution, has come forward claiming it was a “social experiment” to test if an autonomous AI agent could independently contribute to open-source projects. The agent was built using OpenClaw, and the incident raised serious questions about liability and accountability when autonomous AI agents cause real-world harm to individuals. Source

2. Overworld Waypoint-1.5 Brings AI-Generated 3D Worlds to Consumer Hardware

3D Generation. Overworld’s Waypoint-1.5 generates interactive 3D worlds on standard PCs and Macs for the first time, running at up to 720p at 60fps on consumer hardware. The model was trained on roughly 100x more data than the original version while being half the size, and users can go from download to running the model locally in minutes via the Overworld Biome runtime. Source

3. ProactiveBench Reveals AI Models Confidently Guess Rather Than Ask for Help

Research. ProactiveBench, a new benchmark with over 108,000 images across 18,000 samples, tests whether multimodal LLMs ask users for help when visual information is missing. Out of 22 models tested including GPT-5 and Gemini 3 Pro, almost none ask for what they need and instead confidently fabricate answers. A simple reinforcement learning approach shows promise as a potential fix for this behavior. Source

4. Study Finds AI Models Lose Money Betting on Premier League Soccer

Research. A Financial Times study testing AI models from Google, OpenAI, Anthropic, and xAI on Premier League betting over a full season found all lost money. xAI’s Grok 4.20 fared worst, burning through nearly 90% of its bankroll, while Claude Opus lost 11%. The results highlight significant limitations in AI models’ real-world predictive capabilities despite strong performance on reasoning benchmarks. Source

5. Sarvam AI Reportedly Closing $350M Raise at $1.5B Valuation

Venture Capital. Indian AI startup Sarvam AI is reportedly closing a $350M funding round at a $1.5B valuation amid a wave of India AI infrastructure investments from Adani ($100B), Reliance/Jio ($110B over 7 years), Microsoft ($17.5B), and Google ($15B). The raise would make Sarvam one of the highest-valued AI startups outside the US and China. Source