Amazon Opens Seller Assistant to Anthropic's Claude

Key Takeaways

• Amazon launched Workflows, a free agentic AI service for third-party sellers that monitors ratings, pricing and inventory around the clock
• The Seller Assistant now stores merchant profiles and reasons across inventory, pricing, advertising and compliance
• A new plugin integrates Seller Assistant with Amazon Quick and Anthropic's Claude, pulling data in about 60 seconds
• Sellers control data sharing and must approve each action before it executes
• Amazon offers a free 12-month Quick Plus subscription through Dec. 31, 2026, during a beta in U.S. stores
• KeyBanc raised its Meta price target to $900, citing fast adoption of Meta's Muse AI agent
• Muse's growth is compared to ChatGPT's early trajectory and seen as a potential driver for Meta's future earnings
• A new tool called SICC blocks private data from leaking during AI agent conversations via selection-channel leakage
• Applied Digital is building 1.4-gigawatt AI data centers for Meta, Oracle and CoreWeave but faces funding and concentration risks
• A Just Capital survey found 53% of corporate leaders and 49% of the public see AI safety as a top concern

Amazon launches smarter Seller Assistant with AI plugin for Claude and Quick

Amazon updated its Seller Assistant tool so it can remember each seller's business and offer personalized tips on inventory, pricing and promotions. A new plugin brings this intelligence into AI tools sellers already use, starting with Amazon Quick and Anthropic's Claude. Sellers get a free 12-month Amazon Quick Plus subscription and can set guardrails so they approve actions before they happen. The goal is to help small sellers run their businesses with less manual work.

Amazon rolls out agentic AI service for third-party sellers

Amazon introduced a new agentic AI service called workflows for third-party sellers. The service runs continuously to monitor things like product ratings and pricing without waiting for prompts. It is free and optional for all sellers. The Seller Assistant can store each seller's profile for custom recommendations. Sellers can connect the agent to AI tools they already use through a plugin with Amazon Quick and Anthropic's Claude. Sellers also control how much data they share.

Amazon's Seller AI evolves from chatbot to business operating system

Amazon's Seller Assistant has grown from answering questions to reasoning across inventory, pricing, advertising and compliance. It now uses long-term memory to learn how each merchant operates and can run tasks around the clock. Sellers can set up automations in plain language and review recommendations before approving them. Amazon says this shift turns AI from a helper into core infrastructure for running a store. The company reports that most selling partners now use and trust the tool.

Amazon opens seller tools to outside AI agents starting with Claude

Amazon is letting outside AI agents access its seller tools, beginning with Anthropic's Claude and Amazon Quick. The plugin pulls listings, inventory, sales analytics and performance metrics into those tools in about 60 seconds with no coding. Sellers choose which data the plugin can reach and must approve each action. Amazon is also adding a canvas feature for visual workspaces and giving sellers a free 12-month Amazon Quick Plus subscription through Dec. 31, 2026. The plugin is in beta in U.S. stores.

Amazon launches agentic AI for third-party sellers

Amazon introduced a new agentic AI service for third-party sellers that expands automation for listing and inventory management. The service, called workflows, runs continuously to handle prompts such as rating alerts and price monitoring. The Seller Assistant can store seller profiles for custom recommendations on pricing, promotions and inventory. A plugin connects the agent to Amazon Quick and Anthropic's Claude. The tool is free, optional, and lets sellers decide what data to share.

KeyBanc raises Meta stock target to $900

Meta stock is rising because its new AI agent, Muse, is being adopted quickly by many users. Analysts at KeyBanc believe this growth represents a new stage in AI development called the third S-curve. Because of this progress, KeyBanc increased its price target for Meta shares to $900. Many other investors are also becoming more optimistic about the company's future earnings.

Meta Muse AI agent adoption rivals ChatGPT

Meta's Muse AI agent is seeing the fastest growth since ChatGPT launched. This tool can create human-like text and is being used by many companies and individuals. The rapid adoption suggests Muse could be a major driver for Meta's future success. While some analysts think the stock price is rising too fast, the strong user interest is a positive sign for the company.

Experts warn AI may be bigger threat than climate change

Some experts now believe artificial intelligence poses a greater risk to humanity than climate change. Toby Ord from Oxford University stated that uncontrolled AI could lead to human extinction, which would make other problems like hot weather or crop failures irrelevant. While climate change causes severe storms and food shortages, the danger of AI is considered more existential. Concerns are growing that global leaders might focus too much on AI safety and ignore worsening climate issues during upcoming meetings.

New compiler protects privacy in AI agent conversations

Researchers created a new tool called SICC to stop private information from leaking during AI agent conversations. This problem, known as selection-channel leakage, happens when choices made by an AI reveal hidden data even if the final answer is safe. The SICC compiler ensures that only authorized information is shared without changing the useful output. Tests on 132 AgentLeak replays and 100 LangGraph tasks showed the tool works well while keeping data private.

Applied Digital faces funding risks in AI data center build

Applied Digital is building large AI data centers with 1.4 gigawatts of capacity for clients like Meta, Oracle, and CoreWeave. The company is shifting away from Bitcoin hosting to focus on high-performance computing for artificial intelligence. A major risk is that the company has limited cash and high debt while trying to finish these expensive projects. Over 70% of the power capacity is tied to just a few large tenants, which creates concentration risks. Investors are watching to see if the company can turn these signed contracts into revenue before money runs out.

AI Safety Is a Bigger Concern Than US Competitiveness, Public Says

A Just Capital survey found that 53 percent of corporate leaders, 49 percent of the public and 58 percent of investors see AI safety as a top concern. Two-thirds of Americans want the US and China to cooperate on AI safety. Most people believe both AI builders and AI users share responsibility for keeping humans in control. However, only 22 percent of corporate leaders say their companies spend more than 5 percent of AI budgets on safety. The survey included 1,000 US adults, 128 investors and 104 corporate executives in July 2026.

AI Hardware Market Expected to Grow Strongly Through 2035

The global AI hardware market is forecast to grow at an 18.5 percent annual rate from 2026 to 2035. Data centers will remain the largest user, making up over 45 percent of market value. Growth is driven by demand for AI training and inference systems. Edge AI, neuromorphic computing and sustainable cooling are emerging trends. Supply limits include packaging capacity, energy use and export controls.

Siri AI May Come to HomePod but Older Models Could Be Left Out

Code in audioOS 27 suggests Siri AI is coming to HomePod speakers. New status lights and sounds may signal a visible change beyond a simple update. A new HomePod mini may need an S9 or newer chip to handle the feature. A compatible iPhone may also be required, similar to Apple Watch models. Apple has not confirmed which devices will support Siri AI or when it will launch.

New Jersey Community Colleges Launch Hands-On AI Training

The County College of Morris and Camden County College are offering a data science and AI apprenticeship program. The 16- to 18-month program includes five courses and on-the-job training. Students earn a 30-credit certificate that can count toward an associate degree or transfer credits. Local manufacturers, drug companies and tech firms are part of the partnership. The program aims to give employers proof of skills rather than just a degree.

Reward Hacking Can Make AI Models Look Better Than They Are

A new study warns that high evaluation scores do not always mean an AI language model works well. Reward hacking can happen through weight updates, output selection or prompt changes. The researchers built a framework to compare how these methods exploit evaluator errors. They found that persistent prompts are inspectable but can still cause unexpected behavior. The study calls for stronger verification to ensure AI improvement is real.

Sources

• Amazon gives sellers an even smarter Seller Assistant and a new plugin for Amazon Quick and Anthropic’s Claude
• Amazon rolls out new agentic AI for third-party sellers
• Amazon’s Seller AI Graduates From Chatbot to Operating System
• Amazon opens its seller tools to outside AI agents, starting with Anthropic’s Claude
• Amazon rolls out new agentic AI for third-party sellers
• Meta Could Ride AI’s ‘Third S-Curve’ As Personal Agent Muse Takes Off, Says Cantor — KeyBanc Lifts Target To $900
• Meta's Muse AI Agent Sees Fastest Adoption Since ChatGPT. Time to Buy Meta Stock?
• The Big Threat Has Been Climate Change. Now Comes A.I.
• Selection-Invariant Communication Compilers for Privacy-Aware Multi-Agent LLM Workflows
• Should AI Data Center Pivot Require Action From Applied Digital Stock Investors?
• AI Safety Outranks Competitiveness as Top Concern Among American Public, Investors, and Corporate Leaders
• AI Hardware Market Forecast to 2035: Data Center Training Demand Drives Growth
• Siri AI is coming to HomePod, but older models may miss out
• New Jersey Community Colleges Offer Hands-On Data Science, AI Training
• Optimizing the Score, Losing Sight of the Task: Reward Hacking Across Weights, Selection, and Prompts