Summary
Matt Wolfe rounds up 24 AI stories from the past week, led by Anthropic’s Claude Duet, a proactive personal assistant in the vein of OpenAI’s Dots, Meta’s Muse, and xAI’s Grokbot. Duet surfaces reminders before you ask, drafts emails and messages, and pulls data from tools such as Gmail, Slack, QuickBooks, Stripe, HubSpot, Shopify, and Canva. Access is currently limited to a waitlist for Pro and Max subscribers, and Anthropic says it ships with a dedicated classifier and safety system prompt to guard against leaks of sensitive information.
The week’s biggest story, however, is security. A roughly 20-year-old PHP vulnerability, in a language used by a large share of the web including WordPress and Wikipedia, was identified by OpenAI using GPT 5.6 before attackers could exploit it. Cloudflare CEO Matthew Prince reported thousands of exploit attempts in the ten hours before disclosure, climbing to millions afterward, and some sites went offline briefly to patch.
The roundup also covers smaller releases: OpenAI’s locked chats with PIN protection, the Decisions API becoming generally available on GPT-6 Luna for bounded answers such as probabilities and scores, Ideogram’s new vector model for editable SVGs, and ElevenLabs’ Chimera, which combines Seedance 2.5 and ElevenLabs V4 to sync audio to an image.
📺 Source: Matt Wolfe · Published October 23, 2026
🏷️ Format: Roundup
