US order forces anthropic to disable advanced AI models worldwide

The US government ordered Anthropic to disable its Fable 5 and Mythos 5 models for foreign nationals worldwide, citing concerns that users could bypass safeguards designed to prevent identification of software vulnerabilities. Anthropic argued the jailbreak affects only narrow circumstances and does not justify removing models already deployed to large numbers of users. The directive extends US AI controls beyond chips and infrastructure to direct restrictions on model access, creating new tensions between developers and regulators over how model risks should be evaluated.

More than 100 cybersecurity experts and industry leaders urged the government to lift the restrictions. They argued the models' cybersecurity capabilities are not unique and that limiting access could weaken defensive capabilities while doing little to slow adversaries developing similar technologies. The group also called for a more transparent, structured process for evaluating AI-related risks.

Amazon CEO Andy Jassy reportedly raised security concerns about Anthropic's models with senior administration officials before the restrictions took effect. The episode highlights how national-security concerns, corporate relationships, and model-safety debates are becoming intertwined as advanced AI systems move into broader commercial use.

G7 explores trusted-partner access and international AI strategy

G7 leaders discussed a potential trusted-partners framework that would allow approved countries or organizations to access advanced US AI models despite recent restrictions. The proposal emerged after concerns that limits on model availability could undermine international cooperation and reduce confidence in relying on US AI providers. Discussions also covered frontier AI's potential effects on cybersecurity, financial stability, productivity, and labor markets.

Canadian Prime Minister Mark Carney cited the Anthropic restrictions as evidence that countries and organizations should avoid depending too heavily on a limited number of AI providers. He argued that disruptions involving major AI platforms underscore the need for greater diversification across technologies, suppliers, and markets. European Commission Executive Vice President Henna Virkkunen separately emphasized that the EU's AI Act already covers emerging agent technologies and rejected suggestions that European regulations are preventing innovation.

OpenAI expands enterprise deployment, partnerships, and advertising

OpenAI introduced the OpenAI Deployment Company, a business unit designed to help organizations implement AI systems within critical workflows. The initiative includes the planned acquisition of applied AI consulting firm Tomoro and launches with more than $4 billion in backing from investment firms, consultancies, and system integrators. The organization will deploy specialized engineers inside customer environments to identify opportunities, redesign workflows, and accelerate adoption.

OpenAI also launched the OpenAI Partner Network, backed by $150 million and structured around Select, Advanced, and Elite partner tiers. The program aims to certify as many as 300,000 consultants by the end of 2026 and includes a Forward Deployed Experts initiative giving partners access to deployment methods developed through direct customer engagements. Meanwhile, OpenAI is adding advertising capabilities that allow advertisers to generate, modify, optimize, localize, and translate creative using AI, along with expanded measurement features through conversion tracking for app installs and app opens.

New market data shows ChatGPT remains the leading AI assistant globally but its share has fallen below 50% as competitors such as Google's Gemini and Anthropic's Claude gain users. Users are becoming more willing to switch between assistants based on features, ecosystem integration, productivity capabilities, and brand perceptions.

AI search reshapes advertising, content discovery, and commerce

New forecasts suggest AI search advertising could grow from a small share of search revenue today to a substantial portion of the market by the early 2030s. The projections reflect expectations that advertisers will increasingly shift spending toward generative AI experiences as AI-powered search platforms expand their advertising offerings. At the same time, new research suggests that 68% of US Google searches now end without a click to any destination, continuing a long-term trend accelerated by AI-generated search experiences.

AI-referred visitors appear to be more valuable, converting at higher rates and engaging more deeply than traditional visitors. Data cited in a separate report indicates that shoppers arriving through AI assistants convert at higher rates, spend more time on sites, view more pages, and generate greater value per visit than visitors from non-AI sources. Because AI systems increasingly guide consumers through product discovery and evaluation before they reach a retailer's website, many shoppers arrive with stronger purchase intent.

Microsoft is rolling out new AI Performance dashboard features in Bing Webmaster Tools, including Citation Share, Intents, Topics, and Compare. Citation Share shows the percentage of AI citations a site receives for a given grounding query, adding a relative visibility metric alongside raw citation counts. A German court separately ruled that Google can be held directly liable for false statements generated by AI Overviews because the summaries constitute original content created by Google's systems rather than merely displaying third-party search results.

AI agents move into transactions and enterprise automation

Visa and Mastercard are building frameworks to support purchases made by AI agents through tokenized credentials, identity verification, and permission-based transaction controls. Tokenization is evolving from a fraud-prevention mechanism into a broader system for validating authority and permissions within AI-driven transactions. Coinbase launched Coinbase for Agents, a platform that allows AI assistants such as ChatGPT and Claude to execute financial transactions on behalf of users within defined limits, including trading cryptocurrencies, making payments for services, and accessing paid data sources.

Salesforce agreed to acquire autonomous AI agent platform Fin for approximately $3.6 billion, strengthening its Agentforce portfolio. Fin's technology supports customer-service interactions across chat, email, phone, SMS, WhatsApp, and Slack. The deal reflects growing competition among enterprise software providers to establish leadership in AI-powered workflow automation and customer engagement.

Databricks introduced Unity AI Gateway, a platform designed to help organizations manage rising AI costs as agent-based workflows become more common. The system includes spending limits, protections against runaway usage, model-selection recommendations, and tools for monitoring AI consumption across providers. Microsoft is also shifting Copilot Cowork to usage-based pricing and evaluating lower-cost model alternatives as highly active users generate substantial compute expenses.

Why this matters for marketers

The week's developments carry direct implications for marketing strategy, operations, and planning. The US government's restrictions on Anthropic's models signal that access to frontier AI systems can be influenced by export controls and geopolitical relationships, not just by vendor roadmaps. For global enterprises and agencies, this introduces new uncertainty around model availability and international deployment strategies. The growing debate over whether AI jailbreaks can be fully prevented suggests that expectations around safety testing and governance will become more demanding, potentially lengthening timelines between model development and commercial release.

AI search is evolving from an experimental channel into a potentially significant advertising market. The projection that AI search advertising could capture a substantial share of revenue by the early 2030s, combined with evidence that 68% of Google searches now end without a click, means marketers should begin evaluating how AI-driven search experiences fit into media strategies, measurement frameworks, and customer-acquisition plans. The emergence of AI citation metrics in Bing Webmaster Tools gives publishers more structured visibility into how content performs within AI-generated answers.

AI-referred shoppers are proving to be higher-value sources of traffic, converting at higher rates and showing stronger purchase intent. This makes product information, support content, and specifications structured for AI understanding an increasingly important priority. Meanwhile, Visa and Mastercard's progress on tokenized credentials for agentic commerce, along with Coinbase's launch of transactional capabilities for AI agents, suggests that AI Agents & Automation are moving closer to executing actual purchases - not just making recommendations. Marketers may need to optimize experiences for software systems making transactions, not only for human buyers.

OpenAI's expansion into enterprise deployment services and its partner network reflects a broader industry shift: competitive advantage increasingly depends on implementation expertise rather than access to models alone. For AI for Marketing teams evaluating vendors and consulting partners, certifications, integration experience, and deployment capabilities are becoming key differentiators. At the same time, concerns about AI bias, representation, and citation accuracy - highlighted by the KPMG report episode and the GLAAD framework - reinforce the need for strong editorial review processes when AI is used for research, thought leadership, or content production.