Amazon draws a line around AI shopping agents. The U.S. and China talk about an AI hotline. SoftBank borrows billions for OpenAI. And the rest of the industry keeps running.

Picture this. You ask an AI assistant to buy something on Amazon. It finds the product. It compares prices. It is ready to order.

Then it stops. A notice pops up. The agent is "unauthorized." It violates Amazon's Conditions of Use.

That's what happened this weekend to people using Meta's Muse. Amazon says Muse stores customer logins. It scrapes order history. It doesn't say it's a bot. And it never asked for Amazon's permission.

It's a small moment. But it says a lot about where AI is right now. Companies want agents to do our errands. Governments want to manage the risks. And investors are paying more to fund the next big model.

Here's what happened, and why it matters.

Unlock a focused set of AI strategies built to streamline your work and maximize impact. This guide delivers the practical tactics and tools marketers need to start seeing results right away:

Stay ahead of the curve with these top strategies AI helped develop for marketers, built for real-world results.

Download the Free Report

Muse is new. Meta launched it on September 8. It's a personal agent that handles email, calendar, payments, dining and shopping. Within a week, it was the No. 1 free app on iOS.

That made it a real test. Could an AI assistant shop for you inside one of the world's biggest stores?

Amazon's answer was no. The company blocked Muse from Amazon.com. Its reasons: Muse handles sensitive customer data, and it doesn't identify itself as an automated agent. Meta hadn't commented by Sunday night.

This isn't Amazon's first fight like this. It sued Perplexity over its Comet agent. It has blocked similar tools from Google and OpenAI.

The bigger question is simple. When an AI agent sits between you and a store, who owns the relationship?

AI agents promise convenience. They can search, remember what you like, compare options, and reorder the basics. No more ten open tabs.

But that convenience needs access. Logins. Purchase history. Payment steps. The retailer's own systems.

Amazon's move suggests retailers will want a say. How agents identify themselves. What data they can touch. Whether they can buy anything at all.

On Saturday, in New York, Treasury Secretary Scott Bessent sat down with Chinese Vice Premier He Lifeng. He brought an idea. The U.S. and China should tell each other when an AI incident threatens national security.

He called it a move "from opaque to more transparency."

The timing matters. President Trump hosts Xi Jinping at the White House on Thursday, September 25. The two sides also agreed to get a "Board of Trade" up and running. That idea was first discussed in May.

A hotline won't fix the bigger tech rivalry. But it could give both countries a direct line when something goes wrong with AI. Especially when the situation is too serious, or too unclear, for normal diplomacy.

SoftBank owes OpenAI $10 billion. It's the next installment of a much larger commitment. And it's due by October 1.

So SoftBank is borrowing. It has launched a bond sale of about $11 billion. That's $10 billion in dollars and €1 billion in euros. Pricing is set for September 24.

Here's the catch. SoftBank's credit rating is BB+. That's below investment grade. In plain terms, these are junk bonds. Bloomberg says the deal ranks among the largest junk-bond sales ever by a single company. And SoftBank's borrowing costs have been climbing all year.

SoftBank has promised about $65 billion to OpenAI in total.

The AI race used to be about who could build the biggest model. Now it's also about who can pay for it.

The upside

The risk

·       Humanoid robots are still rare

The International Federation of Robotics counted about 7,000 humanoid robots sold worldwide in 2025. Most went to research labs and trial runs. Compare that with roughly 542,000 regular industrial robots installed in 2024. The hype is big. The deployment isn't yet.

·       South Korea lets AI researchers skip the barracks

Starting in 2027, South Korea will let 240 AI researchers do three years of research instead of military service. They need a master's or PhD. Half the spots are at big-company labs. Half are at government and defense research institutes. Big companies are back in the program for the first time since 2013. So far, the first four companies have asked for only 81 of their 120 slots.

·       StepFun previews a 600B-parameter model

Chinese lab StepFun released Step 5 Preview through its API. It costs $1 per million input tokens and $2.70 per million output tokens. It can take in about 1 million tokens at once. StepFun says it will release the weights openly on October 15.

·       Governors want federal AI rules

Maryland Gov. Wes Moore went on CNN on Sunday. His message: states can't handle AI risks alone. He called Trump's planned "AI Force" an "AI farce." Illinois Gov. JB Pritzker made a similar case earlier this month. The debate goes on. Should the rules come from Washington, the states, or the industry itself?

Why did Amazon block Meta's Muse? Amazon says Muse is an unauthorized AI agent that breaks its Conditions of Use. It points to stored logins, scraped order history, no bot disclosure, and no prior permission.

What is Muse? Muse is Meta's personal AI agent. It handles tasks like email, calendar, payments and shopping. It launched on September 8 and quickly became the No. 1 free iOS app.

What is the proposed U.S.–China AI hotline? A way for the two countries to notify each other about AI incidents that affect national security. Scott Bessent proposed it to Vice Premier He Lifeng during talks in New York.

Why is SoftBank selling junk bonds? To fund its next $10 billion payment to OpenAI, due October 1. SoftBank's BB+ rating is below investment grade, so its bonds count as junk.

Are humanoid robots mainstream yet? No. About 7,000 were sold worldwide in 2025, mostly for research and trials.

Go back to that blocked screen. The shopper still wants the product. The agent still knows how to get it. But the store has decided that access isn't automatic.

That same line is showing up everywhere. Between countries deciding what to share. Between investors weighing debt against ambition. Between governors asking for common rules. Between companies deciding who gets to build on their systems.

The next chapter of AI won't just be about what these systems can do. It'll be about who's allowed to let them do it.

Thanks for being a valued subscriber.

Pete Nyandeh

AI Daily Brief, aidailybrief.io

This newsletter is for informational purposes only. Financial figures, product details and policy developments reflect reporting available at publication time and may change. It is not investment, legal or policy advice.