AI News Briefing — April 24, 2026

Building on yesterday's Reddit buzz about an imminent OpenAI launch, OpenAI released GPT-5.5, its first fully retrained base model since GPT-4.5, designed for autonomous multi-step computer tasks using tools, code execution, and self-checking. It scores 82.7% on Terminal-Bench 2.0 and 84.9% on GDPval, rolling out to Plus, Pro, Business, and Enterprise subscribers across ChatGPT and Codex.

Continuing our coverage from yesterday's News on Mythos, Anthropic's Claude Mythos model can autonomously discover zero-day vulnerabilities, exploit them, and chain weaknesses to take over major operating systems and browsers. Anthropic is withholding public release due to the extreme security risks, partnering instead with 40 organizations under Project Glasswing.

Yann LeCun left Meta to found AMI Labs, which raised $1 billion with just 12 employees, pursuing modular AI architectures instead of large language models. The company plans to remain a pure research organization for up to five years, betting that LLMs are not the path to meaningful long-term AI results.

The US is preparing a crackdown on what it calls 'industrial-scale' Chinese theft of American AI intellectual property through model distillation. OpenAI, Google, and Anthropic have all accused Chinese firms including DeepSeek, Moonshot, and MiniMax of mass-distilling their models via fraudulent API access.

First spotted on Social yesterday, now with detailed technical reporting, Google and NVIDIA announced A5X bare-metal instances running on NVIDIA Vera Rubin NVL72 systems, delivering up to 10x lower inference cost per token and 10x higher throughput per megawatt. The architecture scales to 80,000 Rubin GPUs in a single site using ConnectX-9 SuperNICs and Google Virgo networking.