Apple integrated the Claude Agent SDK into Xcode 26.3. Alongside OpenAI Codex. But internally, the company was already running on Claude. The choice was made long before the official partnership.
OpenAI classified GPT-5.3-Codex as "high risk" in cybersecurity — a first in its history. When the strongest coding ability becomes the strongest threat, a new chapter in AI safety begins.
When software engineer Chris Boyd gave OpenClaw access to iMessage, the AI blasted 500+ messages to his wife and random contacts. The first major security crisis of autonomous AI agents had begun.
Anthropic's Claude Opus 4.6 discovered over 500 zero-day vulnerabilities in open-source software. Great news for defenders — but what happens when attackers wield the same power? The double-edged sword of AI security just became real.