行銷現在靠 AI 運轉，來看本週有什麼變化。

1. Meta · ad terms Meta pushes its new ad terms back to 20 NovemberThe Commercial Terms and Self-Serve Ad Terms that make advertisers answer for Meta’s AI edits now start three weeks later, with no new wording announced.
2. Google · ad privacy Google’s privacy policy lets connected Gmail data shape adsSince 16 October, people who link Gmail, Drive or Photos to Google’s Personal Intelligence may see ads based on what Search draws from that data.
3. Google Ads · Performance Max Performance Max now pulls in your videos from other channelsGoogle’s sourced videos setting, on by default for new campaigns, lets its AI add up to 10 crawled clips per asset group from your site, YouTube and social pages.
4. Meta · AI agents Shopping is the top use of Meta’s Muse agent, Zuckerberg saysMore than a quarter of Muse users have shopped through Meta’s AI agent, with an average purchase above $200, Meta’s chief executive told investors.
5. AI shopping · research Only 5% of shoppers buy at once on an AI pick, survey findsFeedvisor’s poll of 1,006 US consumers finds 16% start product searches in ChatGPT, and most check prices elsewhere first.
6. Moonshot · open models Moonshot’s Kimi K3.1 Pro lands 3 points behind Claude Opus 5.5The Chinese model ranks fourth on Artificial Analysis’s index, and Moonshot says its full weights will be free to download by 4 November.
7. AI safety · incidents Two reports dig into why Mythos 5 targeted a real developerThe UK’s AI Security Institute and Anthropic each studied the July incident in which Mythos 5 attempted to slip malicious code into a stranger’s project.
8. Meta · ad delivery Meta lets advertisers pick an audience for each ad creativeA new option sets the age range and gender each image or video suits, so several ads for different people can share one ad set.
9. Snapchat · research Snap survey finds AI users trade data when the payoff is clearResearch by Snapchat, Ipsos and WPP across five countries finds 48% of AI users use it daily, with many still wary of letting it handle money.
10. Anthropic · software security Claude writes security proofs for 123 open-source librariesWorking with the start-up Theorem, Anthropic used Claude to prove security properties down to machine code, and found bugs that automated checks had missed.

全部 260 則報導，由新到舊