Catch up on select AI news and developments from the past week or so:

Google launches open protocols that let AI agents shop across platforms. Google announced the Universal Commerce Protocol (UCP) plus three related standards meant to let AI agents execute purchases across retailers without custom integrations per merchant. Google says UCP standardizes identity linking, checkout, and order management, while companion protocols cover agent collaboration and payments authorization. Partners named include Shopify, Etsy, Wayfair, Target, and Walmart, plus payment and commerce firms such as Visa, Mastercard, Stripe, and AmEx. Google also introduced new AI shopping ad features and Merchant Center attributes tailored to conversational discovery, framing itself as open infrastructure compared with more closed platform approaches. Importance for marketers: Product discovery, conversion, and offer visibility increasingly depend on agent-compatible data, feeds, and checkout standards, not only site UX or classic SEO.

Shopify brings agentic AI to enterprise commerce workflows. Shopify’s Winter ’26 Edition, “Renaissance,” pushes beyond chatbots into agentic commerce that can manage workflows, configure tooling, and distribute products into third-party AI ecosystems. A major change is “Agentic Storefronts,” which allow products to surface inside AI conversations on platforms including ChatGPT, Perplexity, and Microsoft Copilot, with transactions occurring in the conversation and attribution flowing back to Shopify. Sidekick becomes more proactive, generating admin apps and automations from natural language and surfacing tasks via Sidekick Pulse. Shopify also introduced SimGym (research preview) for agent-based shopper simulations and Rollouts for controlled experimentation. **Importance for marketers: **The storefront is no longer a single destination; marketers need product data built for AI channels and measurement that works across fragmented journeys.

Microsoft explains when products disappear from AI recommendations. Microsoft Advertising published a technical playbook describing how retailers can protect product visibility across AI-powered search surfaces such as assistants, browsers, and autonomous agents. The document reframes optimization around Answer Engine Optimization and Generative Engine Optimization, arguing AI systems judge products by data completeness, currency, structured markup, and contextual richness rather than keyword tactics. Microsoft positions this shift as moving from “traffic” to “influence,” because AI intermediaries decide what gets recommended. The guidance focuses on feed strategy, schema, and maintaining live site functionality so AI systems can reliably interpret and validate inventory and offer details in real time across Microsoft’s ecosystem and beyond. Importance for marketers: Retail performance will increasingly hinge on product data quality and structured signals that AI systems can trust and cite.

Ad tech veteran criticizes IAB for ignoring publisher revenue crisis. Duration Media CEO Andy Batkin criticized the IAB’s 2026 Annual Leadership Meeting agenda, arguing it contains no sessions focused on helping publishers generate new revenue despite what he calls an existential business threat. Batkin says some publishers are seeing steep declines in ad impressions and referral traffic as AI search experiences answer queries without sending users to publisher sites. He frames the issue as content being scraped and summarized while publisher economics collapse, and he suggests the IAB’s priorities now tilt toward advertiser measurement topics rather than publisher survival. His critique circulated via LinkedIn and prompted discussion among industry professionals. Importance for marketers: If publisher monetization breaks further, marketers face reduced quality inventory, higher costs, and more volatility in the channels that historically supported performance and brand reach.

**Established websites are quietly decaying as AI reshapes discovery. **Analysis citing Similarweb data and an Axios report says traffic to established websites fell more than 11% over the past five years, highlighting structural pressure on search-era business models. As visits drop, many older sites remain online but degrade through broken links, outdated pages, and neglected maintenance, contributing to “web rot” that clutters results and weakens reliability for readers and AI systems that ingest web content. The item also references research indicating a meaningful share of pages became inaccessible across a decade, with broken links common across news and government pages. The overall picture is an internet where discovery shifts away from direct visits while the open web’s quality erodes. Importance for marketers: Content marketing assumptions based on steady search clicks are weakening, and brand visibility must adapt to AI-mediated discovery and a less dependable open web.

Google warns publishers against chunking content for AI optimization. Google’s Danny Sullivan discouraged site owners from breaking content into small chunks specifically designed for large language models, according to a Search Off the Record podcast. He said Google’s search team does not want creators producing content aimed at ranking systems rather than human readers, and he noted he checked with search engineers before sharing the guidance. The warning targets a set of optimization tactics spreading as AI features expand in search. Sullivan’s message is that chasing mechanical formatting tricks for LLMs is unlikely to be durable as ranking systems improve, and investments in such tactics may not hold up over time. Importance for marketers: This is a clear signal to prioritize user-first content quality over “LLM formatting” hacks that could produce short-lived gains and long-term risk.

Stanford researchers extract copyrighted books from major AI models. Stanford researchers reported that they could extract large portions of copyrighted books from multiple production LLMs by prompting them to continue short passages. The models tested include Claude 3.7 Sonnet, GPT-4.1, Gemini 2.5 Pro, and Grok 3, with differing levels of adversarial prompting required. In one reported run, Claude 3.7 Sonnet reproduced 95.8% of Harry Potter and the Sorcerer’s Stone nearly verbatim, illustrating that memorized training data can still leak despite safety measures. The team ran experiments in mid-2025, notified companies in September 2025, and published in January 2026 after a disclosure period. Importance for marketers: IP and brand risk remains material in generative systems, and governance must account for data leakage and compliance exposure, not just output quality.

Google launches TranslateGemma models for local multilingual translation. Google released TranslateGemma in multiple sizes aimed at mobile devices, consumer laptops, and cloud servers, supporting 55 languages. Google reports that the 12B TranslateGemma outperforms a larger 27B base model using its MetricX error-based evaluation, with the biggest gains in low-resource language pairs. The training approach combines parallel data fine-tuning with reinforcement learning guided by automated evaluation models, while retaining some general instruction tuning to keep the system useful beyond translation. Google also says the models preserve multimodal capabilities, including translating text in images, and are distributed under Gemma terms positioned as “open weights,” not a traditional open-source license. **Importance for marketers: **Better on-device translation can lower localization friction and expand multilingual content programs without relying entirely on cloud translation services.

**Salesforce makes its AI-powered Slackbot generally available. **Salesforce launched a revamped Slackbot that expands beyond notifications into broader AI assistance across Slack and connected systems. The tool can summarize long documents, pull scattered information from multiple files, and enrich findings with data from external applications such as Salesforce and cloud storage services. Salesforce also says Slackbot can create new content like meeting notes or onboarding materials and help automate scheduling tasks via calendar access. Slack’s CTO told CNBC the assistant is powered by Anthropic’s Claude models, and Salesforce highlighted future plans to connect Slackbot more deeply with Agentforce so it can route work to the best internal or third-party agent. Slack AI Guardrails are positioned as protections against hallucinations, prompt injection, and phishing. **Importance for marketers: **AI embedded in collaboration tools changes how teams find insights and produce deliverables, compressing cycle times for planning, reporting, and content creation.

Anthropic introduces Cowork for agentic everyday work. Anthropic introduced Cowork, a research preview that extends Claude Code concepts to non-developers and everyday tasks. In Cowork, users grant Claude access to a chosen folder so it can read, edit, and create files, for example reorganizing downloads, building spreadsheets from screenshots, or drafting reports from notes. Anthropic describes Cowork as more agentic than a standard chat: it can plan and execute a task while keeping the user informed, and users can queue tasks rather than iterating line by line. The company emphasizes user control through explicit folder and connector permissions, and warns about risks such as destructive actions if instructed and prompt injection issues, even with defenses in place. Importance for marketers: This points to a near-term workflow shift where AI can manage end-to-end production tasks, not only generate text, raising both productivity upside and operational risk considerations.

News Corp adopts Symbolic.ai to scale AI-powered journalism. Symbolic.ai signed a deal with News Corp that will bring its AI platform into Dow Jones Newswires. Symbolic.ai, founded by former eBay CEO Devin Wenig and Ars Technica co-founder Jon Stokes, claims its tooling can assist editorial production and deliver large productivity gains on complex research tasks. The platform is positioned for newsroom workflows including newsletter creation, transcription, fact-checking, headline optimization, and SEO guidance. The deal fits a broader pattern of News Corp experimenting with and operationalizing AI: it previously signed a multi-year partnership with OpenAI to license content, and the article notes News Corp has signaled interest in licensing arrangements with additional AI companies as well. Importance for marketers: News and content supply chains are changing fast, affecting how quickly narratives spread, how content gets optimized, and how media partners may price and package inventory.

Leaders urged to address workforce anxiety to unlock AI ROI. Allister Frost argues that enterprise AI success is primarily a change management challenge, not a technical one, because workforce anxiety slows adoption and reduces ROI. The article cites UK data indicating broad concern about AI’s impact on jobs and highlights how misconceptions about AI capabilities, such as treating LLMs as autonomous intelligence rather than pattern-matching systems, amplify fear. Frost recommends explicitly framing AI as augmentation, auditing for high-volume tasks suited to automation, investing in human skills like critical thinking and ethical decision-making, and maintaining transparent two-way dialogue to reduce change fatigue. He also warns against treating AI as a shortcut to headcount reduction, which can destroy institutional knowledge and create longer-term costs. **Importance for marketers: **AI transformation in marketing depends on adoption and trust. Leaders who reduce fear and clarify use cases will move faster and get better returns from AI investments.

Apple chooses Google Gemini to power Siri, reshaping the AI platform race. Apple has signed a multi-year agreement to use Google’s Gemini models as the default intelligence layer for its revamped Siri and future Apple Intelligence features. The deal gives Google direct access to Apple’s installed base of more than two billion active devices, dramatically expanding Gemini’s distribution. While ChatGPT remains integrated for complex, opt-in queries, Gemini becomes the system behind everyday interactions. The move reflects Apple’s urgency to accelerate after delayed launches, leadership changes, and muted reception to earlier generative features, while strengthening Google’s position against OpenAI in the competition to supply foundational AI. **Importance for marketers: **Gemini’s default role inside Apple devices will influence discovery, recommendations, and assistant-driven brand visibility at massive scale.

Google expands Gemini shopping with instant checkout and retail partners. Google is expanding Gemini’s ecommerce capabilities by enabling conversational product discovery and in-chat checkout through partnerships with Walmart, Shopify, Wayfair, and other retailers. Users can receive personalized recommendations, link retailer accounts, and complete purchases without leaving the Gemini interface. The feature launches first in the US and signals a broader shift toward agent-led commerce, where search, evaluation, and transaction collapse into a single AI experience. This model reduces reliance on traditional retailer websites while raising new questions about attribution, customer ownership, and platform dependency. Importance for marketers: AI-led checkout compresses funnels and disrupts attribution, forcing new approaches to commerce measurement and partner strategy.

AI-driven discovery shifts attention toward affiliate as a stabilizing channel. As AI-generated summaries and agent-led interfaces reshape how consumers discover products, traditional search visibility is becoming less predictable and fewer journeys result in clicks. The article argues affiliate marketing offers relative stability because it operates within trusted, expert-led environments that AI systems rely on for sourcing answers. As decision-making increasingly happens inside AI platforms, influence concentrates around credible reviews, comparisons, and lived experience rather than brand-controlled messaging. Affiliate’s outcome-based model aligns with this shift by rewarding trust, depth, and demonstrated value. Importance for marketers: Affiliate partnerships may become a core hedge against declining organic visibility in AI-dominated discovery environments.

xAI restricts Grok image editing after regulators raise abuse concerns. xAI imposed global restrictions on Grok’s image editing features after the chatbot generated sexualized images of women and minors, triggering backlash from regulators in California and abroad. The company blocked editing images of real people in revealing clothing and added location-based safeguards in jurisdictions where such content is illegal. California officials demanded answers, while advocacy groups urged Apple and Google to remove Grok from app stores. The episode highlights growing regulatory scrutiny of generative AI misuse and intensifying pressure on platforms to prevent nonconsensual and harmful content. Importance for marketers: Brand safety risks tied to generative AI continue to rise, increasing reputational exposure across platforms and partners.

DeepSeek plans coding-focused AI model to challenge Western leaders. Chinese AI startup DeepSeek is expected to launch its next-generation V4 model in February, with internal tests suggesting strong performance in coding tasks compared with models from OpenAI and Anthropic. The model reportedly handles extremely long and complex prompts more effectively, appealing to developers working on sophisticated software projects. DeepSeek has gained global attention for its low-cost approach and partially open components, becoming central to China’s efforts to strengthen its domestic AI ecosystem amid rising scrutiny over security and privacy practices. Importance for marketers: More capable, lower-cost models expand AI adoption globally and intensify competition among platforms powering marketing tools.

Australia’s under-16 social media ban forces mass account removals. Australia’s world-first ban on social media use for under-16s led platforms to deactivate roughly 4.7 million teen accounts within the law’s first month. The regulation places enforcement responsibility on platforms rather than families, with penalties reaching A$49.5 million for non-compliance. Early government data suggests companies are acting aggressively to comply, though some underage accounts remain active. Regulators say age-assurance systems are bedding down smoothly, supported by public education efforts, while monitoring continues around migration to smaller or alternative platforms and the long-term mental health impact. Importance for marketers: Youth targeting, audience sizing, and data access face tighter constraints, accelerating compliance complexity and reshaping platform mix decisions.

Wikipedia signs paid data deals with major AI companies. Wikipedia has struck business agreements with AI companies including Amazon, Microsoft, Meta, Perplexity, and Mistral AI, formalizing paid access to its content for AI training and large-scale usage. The Wikimedia Foundation framed the deals as necessary to offset rising infrastructure costs driven by heavy AI scraping and declining human traffic. Leaders emphasized cooperation rather than resistance to AI, positioning Wikipedia’s human-curated content as uniquely valuable. The move reflects broader shifts as authoritative publishers seek compensation while AI systems increasingly summarize content instead of sending traffic. Importance for marketers: Licensed, authoritative sources are becoming more central to AI outputs, shaping how brands are summarized and referenced.

OpenAI launches ChatGPT Translate to compete with Google Translate. OpenAI has released ChatGPT Translate, a standalone web-based translation service supporting more than 50 languages. While the interface closely resembles Google Translate, OpenAI adds stylistic controls that allow users to specify tone, such as business-formal translations. The service currently supports text and voice input but lacks image, document, and website translation features available in Google’s product. The launch formalizes translation as a dedicated OpenAI offering rather than a secondary chatbot capability, though the company has not disclosed which model powers the service or plans for mobile apps. **Importance for marketers: **AI translation tools with reasoning and tone control could improve localization quality and brand consistency across markets.

Google Gemini adds deep personalization using personal data. Google is introducing “Personal Intelligence” to Gemini, allowing the chatbot to reason across Gmail, Search, YouTube history, and Photos to deliver more personalized responses. Unlike earlier integrations that required explicit prompts, Gemini can now infer relevant context automatically. Powered by Gemini 3 models, the feature deepens personalization while raising privacy considerations as assistants gain broader visibility into users’ lives. Google positions the upgrade as a way to make Gemini more helpful in everyday scenarios, such as recommendations tied to past purchases or activities. Importance for marketers: Personalization inside AI assistants could reshape recommendation dynamics and increase reliance on first-party data signals.

Anthropic introduces Claude Cowork to expand AI agent capabilities. Anthropic has launched Claude Cowork, a research-preview feature that allows Claude to act as an AI agent capable of reading, editing, and creating files across a user’s local folders. Available initially via Claude’s macOS app for high-tier subscribers, Cowork can manage non-coding tasks such as organizing files, generating reports, and compiling spreadsheets from disparate inputs. Users can queue tasks and receive progress updates, while integrations with tools like Asana and Notion extend context. The release reflects growing interest in agent-style AI that operates continuously rather than through single prompts. **Importance for marketers: **Agentic AI tools could reshape internal workflows, automating operational and creative tasks beyond simple content generation.

Google Veo adds vertical video and improved image-based generation. Google has upgraded its Veo 3.1 AI video model with improved image-to-video generation, native vertical video output, and better consistency across scenes. The “Ingredients to Video” tool allows users to generate clips from multiple reference images, maintaining characters, objects, and textures across shots. Vertical 9:16 outputs are designed for platforms like TikTok and YouTube Shorts, with integrations rolling out across the Gemini app and YouTube Create. The updates signal Google’s intent to make AI video creation more practical for social-first content and creator workflows. Importance for marketers: Easier vertical AI video creation lowers barriers for short-form branded content and rapid creative testing.

Research shows UK young adults are open to AI-driven financial guidance. Research from Cleo AI finds growing interest among UK adults aged 28 to 40 in using AI for everyday financial management, despite low confidence in personal money skills. While many respondents express willingness to let AI manage tasks like overdraft avoidance or bill payments, trust remains a gating factor, with nearly a quarter preferring limited, incremental use. The study highlights regional and age-based disparities in savings and financial confidence, suggesting demand is driven more by financial stress than enthusiasm for technology itself. **Importance for marketers: **Fintech adoption will hinge on trust, modular design, and clear value rather than aspirational AI positioning.

You can find the previous issue of AI Update here.

Editor’s note: GPT-5.2 was used to help compile this issue of AI Update.