Banks and retailers have started to decide whether an AI agent may act for a customer on their sites. Amazon went to court over it in 2025 and, in September 2026, began blocking Meta’s Muse agent. I have spent the past two years building a multi-agent business operating system that I use every day, and I have been talking with industry leaders about agentic commerce on my podcast. Here is where I think this stands, and what it means for the people who want an agent to handle their errands.
What the courts and the blocks show so far
Amazon filed suit against Perplexity on November 4, 2025, in the Northern District of California. The complaint targets Perplexity’s Comet browser and alleges violations of the Computer Fraud and Abuse Act and California’s computer data access law (Retail Insight Network). It was one of the first cases to ask a court to stop agents from transacting on a vendor’s website.
A district court granted Amazon a preliminary injunction in March 2026. In August 2026, the Ninth Circuit overturned it, finding that on the record before it, Perplexity itself did not access Amazon’s systems (Retail Insight Network). The court limited its ruling to that record, and the case went back to the district court. Amazon has since filed an amended complaint that includes a claim for tortious interference with contract (The Next Web).
On September 20, 2026, Amazon began blocking Meta’s Muse agent. Muse users saw a message stating that continued access by an unauthorized AI agent violates Amazon’s Conditions of Use (The Next Web). Amazon says Muse does not identify itself while browsing and appears to store customer credentials. Meta says Muse has no visibility into people’s passwords or payment methods (The Paypers). I found no lawsuit between Amazon and Meta over Muse, and Amazon declined to comment on possible legal action.
My reading is that the Perplexity ruling turned on anti-hacking law, which leaves a vendor’s terms of service as the main lever for blocking an agent. The Muse block uses that lever.
Why vendors block agents
In my view, the blocks protect more than customers. They protect the vendor’s business model, and I see two clear examples.
When a consumer gives financial data to a personal agent, the agent can optimize spending, consolidate debt and find better alternatives. For a bank, that leads to customers leaving. When a personal agent discovers, compares and recommends products without a human viewing the online store, advertising revenue shrinks, because an agent does not watch ads, click them or respond to persuasion.
Banks and retailers can shut agents out today. I expect that to hold only until a competitor finds a faster way to participate in this new technology economy.
Security concern or profit protection?
Security concerns are warranted at this early stage of agentic commerce. When large organizations gain access to broad sets of consumer data, targeted advertising, spot and surge pricing, and inequality become real concerns.
I would still ask what a given restriction does. Amazon’s complaints about Muse (no self-identification, credential storage) describe how an agent behaves. Amazon’s request that Meta remove Amazon from Muse’s shopping experience (The Paypers) addresses whether the agent may participate at all. A vendor that sets rules for identification and credential handling is protecting its systems. A vendor that bars every agent is also protecting its position.
Governments and consumer protection organizations are better placed than any single vendor to raise these concerns and seek rules, together with self-regulating consortia.
Agentic flooding of customer service and promotions
As consumers adopt personal agents, I expect more of them to look for ways to cut spending, find unused subscriptions and comparison shop without effort. A person who no longer spends hours on the phone negotiating finds it cheaper to try a dispute, a refund request or a request for a better offer.
The organizations on the other end have to absorb the volume. AI tools already make it easier to draft cancellation and service termination letters. What follows is sending those letters to the right person and following up until the matter is resolved. I expect customer service teams to triage these requests with agents of their own, separating legitimate claims from those that are not. Customer service is becoming the next cat-and-mouse game, much like hiring and screening.
Promotions face the same pressure. A coupon or discount aimed at a small audience, or at people who go looking for it, can see far more redemptions once agents do the looking. A calculated promotion then risks selling too much at the discount. It can also disappoint customers when scalpers or second-hand sellers buy the discounted products and resell them at a profit.
What this means for consumers
For now, personal agents already prove valuable on a range of tasks. As adoption grows and companies develop their agentic commerce strategies, more services will open up to them. I expect adoption to follow the early boom of e-commerce and the web: in the near future, every business will transact with agents, just as every business today has a website.
Frequently Asked Questions
Can banks and retailers shut out AI agents that act with a customer’s permission?
They can today, through terms of service and technical blocks, as Amazon did with Meta’s Muse in September 2026. In my view, that works only until a competitor finds a faster way to take part in agentic commerce.
Did Amazon sue Meta over Muse?
I found no lawsuit. Amazon blocked Muse beginning September 20, 2026, citing its Conditions of Use, and declined to comment on possible legal action. Amazon’s lawsuit against Perplexity, filed November 4, 2025, is the case in court.
What is agentic flooding?
It is the surge of requests that customer service systems and promotions face when many consumers use agents to dispute charges, request refunds, cancel subscriptions and redeem offers. I expect companies to respond with triage agents of their own.
About the Author
Andreas Welsch is an internationally recognized AI leader, advisor and speaker who helps enterprises turn technology hype into real business outcomes. He is the Founder & Chief Human Agentic AI Officer at Intelligence Briefing and the 2x best-selling author of the AI Leadership Handbook and The HUMAN Agentic AI Edge.
Hear more conversations about agentic commerce on the What’s the BUZZ? podcast. Andreas is also writing Always-On Agents: What Happens When AI Runs Your Business (and Life), due in January 2027.