AI agents have a big problem: Nobody wants to let them in | usagoldmines.com

Your AI agents will do all your banking, they said. They’ll also shop for you, make dinner reservations, book airline tickets. They’ll do your bidding, all for a low (or not-so-low) monthly fee. Who wouldn’t want an AI agent, right?

Well, it turns out a growing number of businesses want nothing to do with personal AI assistants, the latest being one of the biggest airlines in the world.

United Airlines has declared that AI agents aren’t welcome in the friendly skies, warning that “sharing account credentials with personal, third-party, AI assistants can expose sensitive information, and actions taken by these tools may result in unintended purchases, account changes, or use of MileagePlus miles.”

Starting now, United is “taking steps to restrict access — directly within our digital platforms and indirectly via travel agencies and aggregators — to prohibit use by unauthorized, third-party personal AI agents or assistants.”

Beyond the possibility of an AI agent booking tickets by mistake, they may also “violate FAA and DOT regulations,” adding that checking in for a flight “requires human acknowledgement that bags do not contain banned substances,” United Airlines says.

United isn’t the only company cracking down on AI agents, of course. Late last month, Bank of America sent an email to its customers warning against allowing agents to access their banking accounts. “Please remember that your username and password should never be shared with another person or provided to an outside service, including an AI-enabled tool or agent that offers to act on your behalf,” wrote company executive Ashley Ross in the email (as a BofA customer, I received this in my personal inbox).

Bank of America has (for now) stopped short of banning third-party AI agents from logging into customer accounts, but JPMorgan Chase did, as have Charles Schwab and Citigroup.

And it’s not just airlines and banks slamming the doors on AI agents. Amazon has already made headlines for banning Meta’s Muse agent from its site, while Yelp won’t allow agents that haven’t signed up for a paid partnership. The RealReal and eBay have also reportedly turned their backs on personal AI assistants.

The reasons behind these various AI agent lockouts vary. Some are about protecting the finances of customers, some are to prevent agents from “griefing” other customers competing for the same products, and others are purely about business, with companies banning outside agents in favor of their own AI assistants.

But the common thread is that personal AI agents, armed with credit cards, passwords, and more determination than common sense, are flooding into a world that isn’t ready for them. 

And as online services and businesses hastily set up blockades to deal with the onslaught, the various “killer apps” that the big AI providers were trying to sell us — your agent will buy your plane tickets, book tables at restaurants, handle your banking! — are looking pretty broken right now.

More in AI this week

• No matter what you think of Nvidia or its CEO’s reckless indiscretions, Jensen Huang’s sudden departure has left the AI giant — and, by extension, the US economy — in a perilous state. It’s a stunning development, and who knows what comes next. (New York Times, PCWorld)
• The Jensen-lead Nvidia was the juice behind many AI startups. Now there’s speculation that Jensen’s ouster might “unravel the web of dependencies” holding the AI market together. (Financial Times)
• When I gave my ChatGPT “dot” agent access to my Google Drive, I had no idea it would feel free to proofread my documents and text me its suggestions. (PCWorld)
• Staffers at Proton, a company that offers VPN services and encrypted email, rebelled when CEO Andy Yen ordered them to “stop lovingly handcrafting all our code and use AI more.” (PCWorld)
• Dialing up his IPO rhetoric, Anthropic CEO Dario Amodei is boldly predicting that “things like cancer, Alzheimer’s, heart disease,” and other life-threatening ailments will be “relics of the past” in “less than a decade” if “we do this right.” (Or humanity will be destroyed, take your pick.) (Fireside Alpha)

Prompt of the week: The “surgical edit” prompt

Asking an AI to edit a document can be asking for trouble, as it might go beyond your editorial requests and silently replace your voice with its own. If you’re looking to get ChatGPT or another AI bot to scour a document and make targeting changes, try the “surgical edit” prompt.

Arriving in both long and short versions, this prompt orders the AI to make your edits while leaving the rest of your writing alone. Used with an advanced “reasoning” model like ChatGPT “high” or Claude Opus 5.5, it can be surprisingly effective.

That’s all for now!

Thanks for reading the latest issue of Prompt Mode.

Want more next week? Don’t forget to sign up to start receiving this newsletter in your inbox.