AI News — Friday, April 10, 2026

10
Florida AG Launches Investigation into OpenAI Following ChatGPT-Linked Shooting Allegation

The Florida Attorney General has initiated an investigation into OpenAI after a shooting incident was allegedly linked to the use of ChatGPT, raising questions about AI safety and responsibility.

TechCrunchindustry
9
ChatGPT Finally Offers $100/Month Pro Plan

OpenAI has introduced a new, higher-tier 'Pro' subscription plan for ChatGPT at $100 per month, indicating a move towards premium enterprise features and advanced user access.

TechCrunchproduct
8
Mercor, a $10B-Valued AI Startup, Faces Scrutiny After Data Breach

AI startup Mercor, valued at $10 billion, is under intense scrutiny following a recent data breach, highlighting the growing security challenges for high-profile AI companies.

TechCrunchindustry
8
Google Unveils Veo 3.1 Lite, a Cost-Effective Video Generation Model

Google has announced Veo 3.1 Lite, a new, more cost-effective version of its video generation model, aiming to make advanced AI video creation more accessible.

Google AI Blogproduct
8
Think in Strokes, Not Pixels: Process-Driven Image Generation via Interleaved Reasoning

New research introduces a process-driven approach to image generation that uses interleaved reasoning to create images in 'strokes' rather than pixels, potentially leading to more coherent and controllable outputs.

Hugging Faceresearch
8
RAGEN-2: Reasoning Collapse in Agentic RL

A new paper, RAGEN-2, investigates the phenomenon of 'reasoning collapse' in agentic Reinforcement Learning, highlighting critical challenges in developing robust AI agents.

Hugging Faceresearch
7
OpenAI Outlines 'The Next Phase of Enterprise AI'

OpenAI has published insights into its vision for the future of enterprise AI, signaling strategic directions for integrating advanced AI solutions into business operations.

OpenAI Blogindustry
7
How Well Do Agentic Skills Work in the Wild: Benchmarking LLM Skill Usage in Realistic Settings

Researchers benchmark the real-world performance of LLM agentic skills, providing crucial insights into their practical utility and limitations in diverse scenarios.

Hugging Faceresearch
7
Watch Before You Answer: Learning from Visually Grounded Post-Training

This research explores a novel post-training method for AI models that emphasizes visually grounded learning, allowing models to 'watch' and understand context before generating responses.

Hugging Faceresearch
7
MARS: Enabling Autoregressive Models Multi-Token Generation

A new technique called MARS is introduced to enhance autoregressive models by enabling more efficient multi-token generation, potentially speeding up and improving the quality of AI text outputs.

Hugging Faceresearch
7
Combee: Scaling Prompt Learning for Self-Improving Language Model Agents

Combee presents a scalable method for prompt learning that allows language model agents to self-improve, advancing the development of more autonomous and capable AI systems.

Hugging Faceresearch
7
ClawsBench: Evaluating Capability and Safety of LLM Productivity Agents in Simulated Workspaces

ClawsBench offers a new benchmark for assessing the capabilities and safety of LLM-powered productivity agents within simulated work environments, crucial for real-world deployment.

Hugging Faceresearch
7
INSPATIO-WORLD: A Real-Time 4D World Simulator via Spatiotemporal Autoregressive Modeling

INSPATIO-WORLD introduces a real-time 4D world simulator built on spatiotemporal autoregressive modeling, offering advanced capabilities for virtual environment creation and interaction.

Hugging Faceresearch
6
Unlocking Casual Fun: AI-Powered 'Vibe Coding' for Quick, Niche Apps

A new concept called 'Vibe Coding' leverages AI to enable rapid development of casual, niche applications, making app creation more accessible and fun for developers.

Dev.toproduct
6
Experimenting with AI Subagents

Developers are exploring the concept of AI subagents, breaking down complex tasks into smaller, manageable units handled by specialized AI components to improve efficiency and capability.

Dev.toresearch