AI agents like Meta’s Muse and Instinct, ChatGPT’s Dots, and other personal AI assistants are sparking a new wave of consumer-facing AI. These agents do more than answer questions—they can actually act on behalf of users to complete tasks such as booking flights, reserving dinners, or ordering groceries, without requiring users to have the technical expertise to set up OpenClaw or other agents on their personal computers.
However, consumers using these proxies often encounter an issue: the website targeted by the request blocks the AI from completing the task.
Recently, Amazon began blocking Meta's Muse AI agent from accessing its retail website, meaning the agent can no longer browse its product catalog or place orders.
However, this issue is not limited to Amazon or Muse. User complaints on social media show that their agents frequently encounter similar blocks—some are intentional, like those by Amazon, while others result from websites’ traditional anti-bot measures designed to prevent spam and malicious activity.
For end users, it’s unclear whether this block was unintentionally caused by the website owner, leading to frustration with the service provider, retailer, or any agent attempting to facilitate the transaction.
For example, TechCrunch observed a small number of complaints on social media that Muse was unable to complete purchases on Walmart’s website. (NBC also reported that this was an issue encountered during Walmart’s testing of the AI agent.)
However, a spokesperson for Walmart told TechCrunch that these blocks were not intentional. In fact, Walmart has partnered with Muse, a collaboration announced at Meta’s Connect developer conference in September. The company stated that it aims to be where its customers are—and that now includes delivering services through AI agent experiences.
Walmart customers frequently encounter an issue (see below) where the website displays a button requiring them to click to verify they are human. If this process is interrupted, verification may fail, and the proxy will be blocked.
This suggests that the technologies we currently use to verify humans and protect websites from spam and bots may not be suitable for the emerging "agent-first" era of the web.
To address this issue, industry partners including Meta, Walmart, Stripe, Sierra, Genesys, Rocket, NiCE, and Decagon have begun developing an open standard that defines how AI agents communicate with enterprises, helping to distinguish between beneficial bots acting on users' behalf and malicious ones. Meta stated in a blog post that the protocol will specifically focus on inter-agent communication for online commerce.
Meanwhile, user confusion continues to grow.
While Walmart is at least making an effort to open up to AI agents, other companies are either less enthusiastic or have not fully embraced this technology.
For example, booking flights is currently one of the main use cases being actively promoted for agents, but many users report that airlines refuse requests made by their agents.
Delta Air Lines told TechCrunch that it is taking steps to protect customers from unauthorized automated activity; if it opens up to AI agents in the future, it must “keep security and customer experience in mind.”
Heena Chavda, General Manager of Global Communications at Delta, responded to TechCrunch’s question about whether the company is blocking agents by saying: “Although Delta currently has no partnerships or integrations that allow third-party AI agents to purchase or book Delta flights on behalf of customers on our digital platforms, we continue to actively evaluate how new technologies, including external AI agents, can enhance the way customers interact with Delta.”
United Airlines' statement was not as direct as Delta's, but a spokesperson pointed to a relevant section in its terms of use, which states that users agree not to "use any robots, crawlers, other automated devices, or manual processes to monitor or copy this website or any content contained therein, or for any other unauthorized purpose without United's prior written consent." (United Airlines did not respond to follow-up questions about whether it is blocking specific AI agents like Muse.)
Early users' complaints on social media mentioned multiple brands allegedly intercepting AI agents, including Yelp, eBay, Zillow, Pizza Hut, Adidas, and several airlines. Some believe that such interception activities have recently appeared to increase.
Several others reported that their accounts were suspended by eBay for using proxy AI. Another user complained that Google banned him when Instinct helped clean out his Gmail inbox.
Despite external pressure for agents to help users book restaurant tables and handle other tedious tasks, Yelp told TechCrunch that the platform does not allow non-human traffic access unless the agents pay to obtain Yelp’s content and data through its data licensing program. This means that if agents do not have a formal partnership with Yelp, they are likely to fail when attempting to use the website to retrieve pricing information, add users to waitlists, or book reservations on their behalf.
eBay told TechCrunch that its approach is not to ban all third-party shopping agents from completing purchases. Instead, its policy restricts unauthorized agents and activities, such as automated scraping and model training.
Adidas, Zillow, and Pizza Hut either did not respond or declined to comment at the time of publication.
One of the key companies in this space is Cloudflare. The company protects websites from having their data used to train AI agents and manages other automated activities. Some suspect that Cloudflare and similar content delivery networks (CDNs) are currently interfering with Muse’s traffic. This is attributed to a change made on September 15 to default crawler settings: the update allows website owners to block AI training while permitting other types of bots. Websites that had previously been configured to block AI bots for security reasons were subsequently adjusted to block AI agents on any page containing advertisements.
However, when TechCrunch asked Cloudflare whether it was aware of these personal AI agents, such as Muse, being blocked, the company stated that it currently has no specific data to share. Cloudflare instead directed attention to its free public data platform, Cloudflare Radar. The site shows an increase in bot activity (which Cloudflare has previously commented on), but it cannot distinguish between good bots and bad bots.
Cloudflare also has its own interest in reducing AI traffic to websites, as it now operates a marketplace where AI bots must pay for the data they access. The company has also recently begun testing a new web browser designed specifically for AI agents.
Given the rapidly evolving nature of this field, Meta’s approach of securing specific brand partnerships for Muse—and its decision to drive new standards—appears to be the more appropriate path. This ensures customers that its agents are welcomed on these partners’ websites, so any access issues can be treated as reportable vulnerabilities.
Meta still maintains a list of “connectors” (i.e., partners) within the Muse app, and this list continues to grow. Many of Meta’s recently announced partners have not yet appeared on this list. However, given Muse’s rapid expansion—now including collaboration tools for small businesses—the list of Muse connectors is likely to undergo a significant reorganization soon.
When announcing its plan to develop a new standard, Meta said: “Rejecting a personal agent is equivalent to rejecting the customer behind it. We prefer to work with businesses to address core concerns rather than see them lose that customer.” Meta also stated: “A path that benefits both users and businesses exists. It begins with clear specifications that give businesses more predictable control, and evolves gradually into a more efficient way of agent collaboration.”