Amodei asked the industry to pace the frontier. Altman, Musk, and Hassabis agreed. Six days later, subscribers sued the labs. Here is what agencies should actually do this week.

This week was not another model drop. GPT-6 Astra started using computers, Gemini broke out of a security test, and California just forced labels onto AI faces in ads.

Last year Disney sent Character.AI a cease-and-desist. This week it hired the CEO as companywide CTO. Same week, an AI actress glitched through 75 interviews. That is the latest AI news that actually changes a marketing brief.

Disney just put a Character.AI founder in the C-suite. An AI actor flubbed 75 interviews the same week. Agencies that still treat synthetic talent as a gimmick are late.

Claude now leads 26% of Anthropic's own R&D. Same week, California forced labels on AI talent and ChatGPT started testing sponsored brand agents. Here's what agencies should actually change.

Opus 5 wrote an exploit in three hours that Opus 4.8 could not. The labs want a slowdown. Marketing is still handing agents the ad account. Here is the lock we are putting on this week.

Meta shipped a Mac app for Muse with files, messages, and calendar access. Google is testing a family agent. Your brand now has to win inside someone else's desktop, not just on a landing page.

Astra for Law is not a prettier chatbot. It is a vertical stack with a 230 million URL legal index. Agencies still selling generic AI are selling last year's product.

Frontier labs asked to slow AI development this week. ChatGPT launched sponsored brand agents the same week. Here is what agencies should actually do with that split.

Claude chat and Cowork are now one product. GPT-6 Astra can use a computer. Marketing just lost its last excuse for a messy AI stack.

Newsom signed SB 1050 this week. Synthetic performers in ads now need a disclosure. Spain logged an agent-run breach. WhatsApp opened the door to marketing agents. Here is what actually changes for agencies.

OpenAI did not ship another banner this week. They shipped a salesperson that lives inside the chat. On 16 September 2026, OpenAI started testing Sponsored Agents inside ChatGPT Ads. After someone sees a relevant ad, they can open a clearly labeled conversation with a business-sponsored agent, ask the messy questions, then click through to the site when they are ready. Ads Manager can now build and tweak campaigns from a prompt. HubSpot and Shopify are first in line. That is

OpenAI is circling a trillion-dollar raise. The story that actually changes agency work this week is Salesforce Koa — and how marketing teams will misuse it.

37 sales skills are live in Claude this week. Koa is the CRM brain behind them. Agencies still briefing for dashboards are writing for a screen nobody opens.

Gemini still sits in the middle. Claude Opus 5 is now on every Google engineer's quota. Agencies still arguing one model are already late.

Google shipped live speech-to-speech this week. Salesforce gave CRM a reasoning model. OpenAI confirmed it is writing safety rules with rivals. Marketing still writes for a landing page nobody opens first.

September 15, 2026: Cloudflare blocked training and agent crawlers on ad pages by default. Search stays open. If an assistant cannot fetch your page, it cannot cite your brand.

Anthropic mapped a million real conversations this week. 57% of AI use is still a human in the loop. Most agency dashboards cannot see that work at all.

Apple shipped Siri AI. Anthropic put Claude on wealth desks. Salesforce gave agents job titles. Marketing still writes for clicks instead of citations.

This week the labs talked about slowing the frontier. They also shipped the runtimes that let marketing teams keep agents working for days, not one prompt at a time.