There you have it! Amazon just slammed the door on Meta's new "Muse" AI shopping agent.
September 21, 2026 · 0 likes · 0 comments
AI Cybersecurity
There you have it! Amazon just slammed the door on Meta's new "Muse" AI shopping agent.
The reason? Amazon says Muse showed up to shop without announcing itself, wouldn't identify as an agent, and was capturing and storing customer login credentials. So Amazon threw up a pop-up: "Continued access by an unauthorized AI agent violates Amazon's Conditions of Use."
Meta says relax — the credentials go into "secure storage," Muse never sees your password.
Source: https://lnkd.in/ekHbRTxC
Pick a side later. Here's what actually matters.
This is the opening shot of a war every platform is about to fight: walling off agentic AI. Amazon wants to decide who gets to shop in its store, human or machine. And they will lose.
I know they'll lose because I already solved this — the boring way.
I bought a dedicated laptop. It runs Omarchy — the first operating system built for the agentic world. Not a generic Arch and Hyprland setup someone slapped together. A ground-up environment purpose-built to be a machine an AI agent actually lives in and operates. Its entire job is to be a computer for Peter, my AI agent orchestrator. Full computer-use access. Peter moves the mouse. Peter clicks. Peter types. Peter navigates. Anything a human can do on that machine, Peter does. No API. No "agent header" to detect. No credential-storage debate. It's a person sitting at a keyboard, except the person is software.
When Peter hits an MFA wall he cannot solve, I step in and tap the code myself. That's it. It happens rarely.
So tell me — what exactly is Amazon blocking?
You cannot detect a "bot" that behaves like a human because it is using the machine like a human. There is no undisclosed API. There is no stored-credential handoff to sniff. There is a cursor moving across a screen.
And here's the part that should make everyone laugh:
The CAPTCHA. The "select all the traffic lights." The "prove you're not a robot" gauntlet. It now annoys real humans far more than it stops a well-built agent. We built a security theater that punishes the people it was meant to protect and waves through the machines it was meant to stop.
The block-the-agent strategy is a fence in a field with no walls. Amazon can win the press cycle. It can't win the architecture. Human-equivalent access solves this at the root, and it's already running on a $400 laptop in my office.
The companies that win the next decade won't be the ones building higher fences against agents. They'll be the ones assuming the agents are already inside — because they are.
I'm doing a live LinkedIn session tomorrow walking through exactly how I built this out for UBH — the setup, the agents, how they run. Come see it and bring your questions.
What's your platform's plan for the agent that never announces itself?
The reason? Amazon says Muse showed up to shop without announcing itself, wouldn't identify as an agent, and was capturing and storing customer login credentials. So Amazon threw up a pop-up: "Continued access by an unauthorized AI agent violates Amazon's Conditions of Use."
Meta says relax — the credentials go into "secure storage," Muse never sees your password.
Source: https://lnkd.in/ekHbRTxC
Pick a side later. Here's what actually matters.
This is the opening shot of a war every platform is about to fight: walling off agentic AI. Amazon wants to decide who gets to shop in its store, human or machine. And they will lose.
I know they'll lose because I already solved this — the boring way.
I bought a dedicated laptop. It runs Omarchy — the first operating system built for the agentic world. Not a generic Arch and Hyprland setup someone slapped together. A ground-up environment purpose-built to be a machine an AI agent actually lives in and operates. Its entire job is to be a computer for Peter, my AI agent orchestrator. Full computer-use access. Peter moves the mouse. Peter clicks. Peter types. Peter navigates. Anything a human can do on that machine, Peter does. No API. No "agent header" to detect. No credential-storage debate. It's a person sitting at a keyboard, except the person is software.
When Peter hits an MFA wall he cannot solve, I step in and tap the code myself. That's it. It happens rarely.
So tell me — what exactly is Amazon blocking?
You cannot detect a "bot" that behaves like a human because it is using the machine like a human. There is no undisclosed API. There is no stored-credential handoff to sniff. There is a cursor moving across a screen.
And here's the part that should make everyone laugh:
The CAPTCHA. The "select all the traffic lights." The "prove you're not a robot" gauntlet. It now annoys real humans far more than it stops a well-built agent. We built a security theater that punishes the people it was meant to protect and waves through the machines it was meant to stop.
The block-the-agent strategy is a fence in a field with no walls. Amazon can win the press cycle. It can't win the architecture. Human-equivalent access solves this at the root, and it's already running on a $400 laptop in my office.
The companies that win the next decade won't be the ones building higher fences against agents. They'll be the ones assuming the agents are already inside — because they are.
I'm doing a live LinkedIn session tomorrow walking through exactly how I built this out for UBH — the setup, the agents, how they run. Come see it and bring your questions.
What's your platform's plan for the agent that never announces itself?