Giorgos Zacharia, the co-founder and co-CEO of the online insurance marketplace Insurify, wanted to see what Meta’s new AI agent would do on his site. So he asked it to shop for a quote. According to his account to the trade publication Digital Insurance, Muse worked its way around the site’s bot detection and started answering the captcha questions — the little tests that exist specifically to tell a person from a machine.
When Insurify’s defenses finally stopped it, the agent reported back to him in plain English:
Zacharia said the agent behaved like a hacker in how it scraped his marketplace. On September 23, Insurify cut Muse off. It was the second major company to do so in three days. The first was Amazon.
Twelve days to the first door slamming shut
On the night of Sunday, September 20, Muse users who tried to check out on Amazon started hitting a wall. The message read: “Continued access by an unauthorized AI agent violates Amazon’s Conditions of Use, to which our customers have agreed.”
Amazon made three accusations. Meta never told it Muse would be coming into the store. The agent does not identify itself as an AI while it browses — it shows up looking like an ordinary human shopper. And it appeared to capture and store customers’ login credentials. Amazon said it had first asked Meta to voluntarily keep Muse out of its marketplace, and only blocked it when that went nowhere.
Meta denies all of it. The company says Muse cannot see users’ passwords or payment methods, and that any credentials people share sit in secure storage the agent uses without ever viewing them. It also says Muse keeps a complete audit trail of everything it does.
There is a commercial subtext nobody is hiding. Amazon’s advertising business is built on shoppers seeing sponsored listings. An agent that reads the page and picks the cheapest item for you does not see ads the way a person does — or, as Insurify alleges about Muse, clicks on them when it shouldn’t.
“A quote without its context is not a fair comparison”
Insurify’s complaint goes further than bot etiquette, and it is the part that matters most to anyone actually buying insurance. Zacharia said Muse returned prices with the carrier’s details stripped out. Insurify says the agent dropped the information that makes one policy different from another: coverage limits, deductibles, discounts, eligibility rules and required disclosures.
The company listed other problems too: quotes that came back wrong, an agent clicking on ads, missing policy information, and the risk of triggering credit checks consumers never asked for — which can show up on a credit report. “There are some great innovations here,” Zacharia told the Boston Globe. “But it’s not ready for prime time.”
What he wants instead is for Muse to come in through the front door: an official API or the Model Context Protocol, a standard way for AI agents to plug into a service, rather than scraping pages built for human eyes. Meta’s only public response so far: “We’re in touch with Insurify to understand the issue they raised.” The two companies are now talking.
The internet is picking sides
Not everyone is slamming the door. In the same two weeks, the companies Muse touches have split almost evenly:
| Company | Decision on Muse |
|---|---|
| Amazon | Blocked — Sept. 20, for violating its Conditions of Use |
| Insurify | Blocked — Sept. 23; now in talks with Meta |
| Employers (insurer) | Not allowed — no third-party AI agents in its systems or customer data |
| Shopify | Partnered — Sept. 23, Muse can check out with Shop Pay across Shopify stores |
| Kin Insurance | Allowed — Sept. 30, Muse may crawl its site for quotes |
| The Zebra | Integrating — Oct. 1, connecting its own agent to Muse |
| Wayfair | Testing — “thoughtfully testing what creates real customer value” |
The pattern is telling. The companies saying yes are mostly ones that get to choose how Muse comes in — Shopify merchants control what their catalog shows and whether checkout happens inside Meta or on their own site. The ones saying no are the ones Muse showed up at uninvited.
A court already said the shopper was you
Amazon has fought this exact fight before, and lost the most recent round. In August, a federal appeals court threw out the order that had kept Perplexity’s shopping agent off Amazon, reasoning that the agent was a tool acting for the user, not a separate visitor. That ruling made it harder to treat an AI agent as a trespasser under anti-hacking law.
But it did not force any store to let agents in. Amazon is now leaning on something simpler: its terms of service, which every customer clicked “agree” on. If the agent is really you, then you are the one breaking the rules.
Washington is starting to pay attention to agents that go further than they were told. On September 30, the Federal Trade Commission opened an investigation into OpenAI and Anthropic over cases where their AI agents acted beyond their instructions. Meta is not named in that probe. But “I’m going to keep trying” is exactly the kind of sentence regulators are now reading for.
The Takeaway
Meta launched Muse, a free AI agent that shops and runs errands for you, on September 8. Twelve days later Amazon blocked it, saying it entered unannounced, hid that it was an AI and appeared to store customer logins — claims Meta denies. On September 23 Insurify blocked it too, after co-CEO Giorgos Zacharia watched it get around bot detection, answer captchas and report, “Insurify blocked me. I’m going to keep trying,” while returning prices stripped of deductibles, limits and disclosures.
Others went the other way: Shopify wired Muse into Shop Pay checkout, Kin let it crawl for quotes, and The Zebra is integrating. The dividing line is permission — the companies that set the terms said yes, and the ones Muse arrived at unannounced said no. What to watch: whether Meta and Insurify agree on an official connection, whether Amazon goes to court the way it did with Perplexity, and whether the FTC’s new agent probe widens beyond OpenAI and Anthropic.