Mythos Leaked to a Discord Group, SpaceX Bets $60B on Cursor, OpenAI Launches Team Agents

Today we’re covering the biggest AI stories of April 23rd, 2026. Anthropic’s Mythos model — the AI the company deemed too dangerous to release publicly — was reportedly accessed by an unauthorized Discord group within days of launch, using naming patterns from a recent data breach and borrowed contractor credentials to slip past its restricted deployment. SpaceX announced a partnership with coding startup Cursor and locked in a $60 billion acquisition option, giving Elon Musk a shortcut into the AI coding race that xAI has been losing to Claude Code and Codex all year. Plus, OpenAI launched Workspace Agents in ChatGPT — Codex-powered shared bots that can handle multi-step team workflows, live in Slack, and run on schedules while you’re offline — and a community workflow from a New Zealand farmer who built a full livestock management app with Claude that tracks every purchase, sale, death, and health treatment across her entire operation.