What should have been ordinary billing fixes turned into costly delays for some Anthropic customers, who say the company's AI-driven support kept routing them away from a human agent.
For some Claude users, the fallout led to duplicate charges, card disputes, and even paying for extra subscriptions just to keep work moving.
Here's what to know
In those accounts, billing and refund issues often stayed unresolved for weeks. Six Anthropic customers told The San Francisco Standard they ran into billing or refund problems that stretched on for weeks, and only two ultimately got a response from a human, both over email.
Anthropic's support flow runs through an external AI helpdesk product. The outlet reported that Anthropic uses Fin, an AI customer service platform that Salesforce acquired in September for $3.6 billion. In a testimonial on Fin's site, Anthropic said the system handles about 560,000 customer service resolutions each month, The Standard reported.
Tom, 33, who runs a startup sourcing computer parts from his Bay Area home, said that when he tried to upgrade his Claude subscription, charges kept hitting an old credit card.
"These things aren't deterministic systems, so sometimes [the AI agent] can fail and just decide on its own that it doesn't want to patch you through to a human," Tom said, per The Standard.
TCD Picks » Quince Spotlight
💡These best-sellers from Quince deliver affordable, sustainable luxury for all
Consumer advocates say that kind of unpredictability creates a transparency problem. Justin Kloczko, a policy researcher at the nonprofit Consumer Watchdog, told The Standard, "One person's refund is completely different from another customer's journey, even when they're buying the same exact thing, because it's based upon data or what the AI perceives."
More background
Anthropic's support system appears to rely heavily on automated judgment calls.
Fin lets clients feed in user data and conversations, then create rules for when a case should be escalated. The chatbot also detects "sentiment," including whether a user appears "frustrated" by repeating a request, complaining about a charge, or threatening a bad review.
Fin's pricing structure may also encourage keeping customers with the bot. Fin's website, as described by The Standard, says Anthropic is charged 99 cents when an AI chatbot resolves a case and nothing if the issue is escalated to a person.
David, a tech professional in his 30s, said that a mistaken $200 annual plan charge led his bank in Taiwan to suspend his card during the dispute, creating problems for other bill payments, according to The Standard.
While AI tools can help businesses and utilities operate more efficiently, they depend on energy-intensive data centers that can consume large amounts of electricity and water. They also raise concerns about security, misuse, and unintended costs that can ripple out to households and the grid.
What's being done?
California has already passed a measure aimed at this kind of automated dead end.
Gov. Gavin Newsom signed AB 1609, requiring businesses with more than $500 million in annual revenue to make a good-faith effort to connect customers to a human agent within 15 minutes of a request or to schedule a callback within one business day. Penalties can reach $5,000 for a first violation and $10,000 for later ones.
Not every business covered by the law must create a new phone support channel. The law also includes a carveout: A business without a phone support line in place by Jan. 1, 2027, does not have to add one if it offers another prompt way to reach a customer service agent.
Still, Assemblymember Rick Zbur, who authored the bill, said the 15-minute escalation rule would apply even if AI is handling the first layer of support through chat.
"It will result in companies having to staff differently and actually design their systems differently than they do today," Zbur said, per The Standard.
Some Anthropic customers told The Standard they are keeping backup subscriptions, searching social media for direct contacts, or considering chargebacks — even though some Claude users say disputes can lead to account shutdowns.
"If an AI customer service system is not fully predictable or controllable, I would definitely want there to be a clear way to reach a human representative. A service with no real human support makes me feel much less comfortable, especially when it involves billing or refunds," David said.
Get TCD's free newsletters for easy tips, smart advice, and a chance to earn $5,000 toward home upgrades. To see more stories like this one, change your Google preferences here.
