AI agents are finding the web has many locked doors
AI agent
Software that performs online tasks for a person.
anti-bot system
A website check meant to stop unwanted automated programs.
data licensing
A paid agreement that allows someone to use a service’s information.
What happened
The Verge reported on October 17 that AI agents are hitting roadblocks across popular websites. An AI agent is software that can act online for a person. It can search, book a table, arrange travel, or buy products. Users of Meta’s Muse, xAI’s Grok, OpenAI’s Dots, and Instinct reported that their agents could not complete tasks on major sites.
Yelp told TechCrunch that it permits agents only when they pay for access through its data licensing program. Delta Air Lines does not currently have a partnership or integration that lets a third-party agent book flights for a customer. JPMorgan Chase and Charles Schwab confirmed that they block agents from accessing accounts because of security concerns. Reports about Cars.com and Amtrak came from users, so they are not the same as statements from those companies.
The story’s Hacker News listing received 110 points and 93 comments. Those numbers measure attention from that community. They do not prove that the original reporting is correct.
Why sites are blocking agents
Some blocks are deliberate. Others may come from older anti-bot systems. Websites use these systems to fight spam, fraud, and harmful automated activity. An agent acting for a customer can look similar to an unwanted program. That can cause a block even when the website does not intend to reject the customer.
Amazon has already fought Perplexity over its agentic shopping feature. Amazon banned the feature and sued the startup. A judge ordered Perplexity to stop shopping on Amazon, but an appeals court overturned that order in August. Amazon has since filed an amended complaint. The dispute raises practical and legal questions about whether an agent is a customer’s assistant or an unauthorized automated visitor.
Cloudflare has also shaped the environment. It introduced a one-click block for AI crawlers in 2025. It later blocked crawlers that scrape for search indexing and AI training at the same time. The Verge connected that wider effort to the problems agents face, but Cloudflare said it had no specific data about personal-agent blocks.
Why it matters
The usefulness of an agent depends on more than its model. The website on the other end must accept its access and actions. A capable agent still cannot book a flight if the airline refuses the connection. If every website applies different rules, users may get a patchwork of services instead of one broad web assistant.
The failure is also hard to explain. A customer may not know whether the agent failed, the website blocked it, or an anti-bot check misread it. That uncertainty can weaken trust in both the agent and the service. Security protections remain important, but broad defenses can also keep helpful software out.
What is confirmed
The company responses show several different approaches. Yelp points agents toward paid data access. Delta has no current third-party booking integration. JPMorgan Chase and Charles Schwab block account access. Meta is building a permitted lane through partnerships with Walmart, Gap, Sephora, Best Buy, and other retailers. Shopify has opened its storefront network to agents.
The Verge asked OpenAI and Google about the roadblocks facing their agents. It did not receive an immediate response. Instinct declined to comment. Those facts show that the companies have not presented one shared access policy.
What remains unclear
It is not clear how many reported blocks are intentional. It is also unclear how often older anti-bot defenses cause accidental failures. User reports about Cars.com, Amtrak, Google accounts, and Walmart do not establish a single industry-wide rule. Cloudflare’s exact role in blocking personal agents is also unresolved.
The Amazon-Perplexity case is still moving through legal channels. It remains uncertain whether partnerships, paid access, or a new standard will become the normal way for agents to use the web.
What to watch next
The important signals will be official partnerships, licensed data access, and clearer rules for agent traffic. Watch whether websites can tell helpful agents from harmful automation. Also watch whether they explain failures clearly to users. The next stage of AI-agent competition may depend as much on permission to enter the web as on the ability to act once inside.
Why many websites are blocking AI agents
📰 Full story: AI agents are finding the web has many locked doors
AI agents can do online tasks, but websites do not always let them in.
AI agent
Software that does online jobs for a person.
anti-bot system
A website check that tries to stop unwanted computer programs.
data licensing
Paid permission to use a service’s information.
💡 The gist
- AI agents can do online tasks for people.
- More websites are blocking or limiting those agents.
- Hacker News attention does not prove the report is true.
The Verge, a technology news site, reported this change on October 17. An AI agent is software that does online jobs for a person.
Meta, the company behind Facebook, has Muse, a personal AI assistant. OpenAI, the company behind ChatGPT, has an agent called Dots. Users also reported blocks affecting xAI’s Grok and Instinct.
Yelp said agents must pay for access through its data licensing program. Delta Air Lines has no current partner system for third-party agents to book flights. JPMorgan Chase and Charles Schwab block agents from customer accounts for security reasons.
Some reports mention Cars.com and Amtrak. Those reports came from users. They are not the same as official company statements.
Why does this happen? Websites already use an anti-bot system. It tries to stop spam, fraud, and harmful programs. Helpful agents can look like unwanted bots. The website may block the agent without meaning to block the customer.
Amazon blocked Perplexity’s shopping agent and sued the startup. An appeals court overturned an earlier order in August. The legal dispute continues.
Meta is choosing a different path. It partnered with Walmart, Gap, Sephora, Best Buy, and other retailers. Shopify also opened its store network to agents. These partnerships give agents an approved way to enter.
Cloudflare helps websites control automated visitors. It introduced a one-click block for AI crawlers in 2025. The Verge says this wider effort may affect agent access. Cloudflare has not shared specific data about personal-agent blocks.
The Hacker News listing received 110 points and 93 comments. That shows community attention. It does not prove every detail in the report.
The main lesson is simple. A smart agent still needs permission to act. In the future, some websites may welcome agents. Others may require a partnership or paid access. Users will also need clear explanations when an agent fails. The web is becoming a place where every site can set its own rules for AI helpers.
💬 Why Some Sites Block AI Agents
AI agents can make searching and shopping easier, but they may also remove ad revenue, create server load, and trigger false alarms against humans. The future is still uncertain.
- Many sites make money from ads and human attention. An agent may skip the ads and choose products directly, which could hurt that income.
- An agent may compare price, shipping, delivery, and returns. But not everyone will use one: some people distrust them, cannot afford the service or hardware, or do not need one often.
- Sites also worry about heavy bot traffic. One commenter self-reported nginx logs dropping from about 1 GB a day to about 10 MB a week after blocking bots. Another user self-reported being blocked by DataDome despite being a human.
- One proposed answer is a paid, approved API or feed for agents. Yet that feed could still contain ads, marketing manipulation, or prompt injection.
- The main disagreement is whether websites should protect their business and servers first, or let a user’s agent browse freely. There is no settled answer.
initial digest at 93 comments (revision 1). We fetched 95 comments and sampled 95 across the thread. These are HN users’ reports, not independently verified facts.
AI helpers are meeting locked websites
📰 Full story: AI agents are finding the web has many locked doors
Computer helpers can do jobs online, but some websites say no.
AI agent
A computer helper that does internet jobs for a person.
AI agents are computer helpers. They can do internet jobs for people.
The Verge is a technology news site. It told this story.
Meta is the company behind Facebook. Its helper is called Muse.
OpenAI makes ChatGPT. Its helper is called Dots.
Some helpers cannot enter websites.
Yelp lets in helpers that pay for its information.
Delta Air Lines does not have a way for helpers to book flights.
JPMorgan Chase and Charles Schwab keep helpers out of money accounts.
Websites use anti-bot systems. These systems try to stop bad computer programs.
Good helpers can look like bad programs. Then the website may stop them.
Amazon stopped Perplexity’s shopping helper. Meta works with some stores. Shopify lets helpers use its stores.
Hacker News showed 110 points and 93 comments for this story.
That shows people noticed the story. It does not show that every detail is true.
Some websites may welcome helpers. Other websites may still say no.
💬 Websites and Shopping Robots
An AI agent is like a little robot that looks for things for you. Websites may still want to stop it.
- The robot can compare prices, shipping, delivery, and returns. But it may skip ads, and ads help many websites pay their bills.
- Not everyone will trust or pay for a robot. So the idea that every person will use one is only a guess.
- Some users say bots can be heavy: one reported nginx logs falling from about 1 GB a day to about 10 MB a week after blocking them. Another said anti-bot software stopped a real person from signing up.
- A paid, approved door for robots might help. But ads and tricky information could still get through that door.
initial digest at 93 comments (revision 1). We fetched 95 comments and sampled 95 across the thread. These are HN users’ reports, not independently verified facts.
💬 Why Websites Are Closing Their Doors to AI Agents
HN commenters describe a conflict between websites funded by human attention and agents that act for users to find the best option. They also stress that the prediction of agents becoming everyone’s main interface is still only a prediction.
initial digest at 93 comments (revision 1). We fetched 95 comments and sampled 95 across the thread. These are HN users’ reports, not independently verified facts.