ewsba.se
United Airlines said Wednesday it would be restricting artificial intelligence bots from its digital platforms, warning that allowing third-party AI assistants to access account credentials could expose sensitive personal information, lead to unintended purchases or other unauthorized account activity. The move is similar to Amazon's recent blocking of Meta's Muse agent from making purchases on its site.
Still, AI agents can complete a wide range of tasks on your behalf. Instead of just asking Muse what the weather will be tomorrow, you can tell it to suggest an outfit based on tomorrow's weather forecast and book you a rideshare if it's going to rain, for example. Or, rather than just creating an itinerary for your next vacation, you can connect a payment option and have your Dot — OpenAI's agent offering — book flights, hotels and excursions for you.
You can even have an AI agent buy and sell stocks through investing platforms like Robinhood.
Relying on an AI agent for any of these tasks may come with mixed results depending on which agent you use, what data you allow it to access and the complexity of the task you've asked it to carry out. Because broadly, agentic AI is still in its infancy, you may not want to let an agent make decisions "anytime [your] health, wealth, or privacy are involved," says Shomit Ghose, adjunct professor at the University of San Francisco School of Management, and lecturer at the University of California, Berkeley College of Engineering.
"You can have agents do interesting things for you, like categorize your spending, or flag unusual charges, or model what-if scenarios," he says. "Those are good things, but bad things would be moving money, making trades, filing forms. All those things should stay manual at this point."
AI-enthusiast traders may need some convincing, though.
Here's what you should consider before outsourcing your whole to-do list to an AI agent.
Though AI agents' capabilities are improving, they still operate in a fairly literal sense, which can create problems, Ghose says. If you tell your agent to book you the cheapest flight, for example, it may book you a flight at 3 a.m. or one with three layovers, unless you give the agent specific guardrails.
"The bot only knows your context," Ghose says. "It cannot infer, as we humans are able to do, because of instinct and culture and education."
Further, bots can still get it wrong even with a seemingly clear and direct command. They "treat all data as being equal," Ghose adds. So if it finds information on the web that is either inaccurate or subjective, it could give you misleading guidance or fail to meet your standards for the given task.
For example, if you tell your agent to book you a reservation at a good barbecue restaurant, it may get you a table at a place that claims to be "the best barbecue in the world" on its website. But the restaurant may be in an inconvenient location or have poor reviews elsewhere on the web. Your command may need to be more specific, like asking your agent to check reviews on Google Reviews or Yelp.
Depending on what task you've assigned the bot, the outcome could be rather innocuous. But when individuals consider letting bots make purchases for them, make changes to their investment portfolios or and the like, the result could be more consequential.
"In the end, you want to ask yourself: Does the human need to be in the loop?" Ghose says.
If you're booking a vacation, for example, it's reasonable to ask an AI agent to help make a decision on your destination based on the weather, your budget and other factors. But actually booking the flights and hotels "should still come back to me," Ghose says. He wants to maintain final say over major financial decisions.
Doing so may even save you money. In Booking Holdings' third-quarter earnings report, the Connecticut-based travel technology company reported 20% higher spending among customers who used AI assistants versus those who didn't to book accommodations on Agoda, one of its travel platforms. It's unclear whether those customers were led to pricier options or did, in fact, seek more expensive amenities. The company doesn't cite a single factor for the finding.
AI agents are also making their way into the healthcare sector, which can pose risks for users too quick to rely on the agents' advice, Ghose says. Letting an AI agent handle prescription refill requests, for example, could be convenient, but AI managing prescriptions more broadly may act on "incomplete records and won't be able to recognize a dangerous interaction or a worsening condition before the action has already taken effect," he says. Generally, a clinician will have to sign off on any new prescriptions, but federal agencies are backing programs that would allow AI agents to assist in diagnostics and prescription management for patients.
Relatedly, anytime you grant an AI agent access to your personal information, whether it's medical records or your email inbox, you could be introducing privacy risks, Ghose says. He recommends users only allow agents to "use limited permission accounts," like letting it access your subscription to a news source, rather than your personal accounts, such as your bank account or credit card information.
For some tasks like investing decisions, you need to have a certain level of expertise to clearly dictate your strategy and goals. And even if you give yourself final say on trades, instead of delegating to the bot, it's essential that you understand the risk involved in moves your AI agent proposes. While a large language model can help you research investments and gather information around an asset, humans are still better at putting that into the context of their individual financial strategy and goals, AI experts say.
Agents can "make your trading experience easier, there is no doubt about it," Shridhar Rathore, an experienced investor who uses agentic trading tools, recently told CNBC Make It. "But [letting agents] make investment decisions, that's where people can lose money if they are not careful...they have to make sure that for the execution, the human is in the loop."
"You want to be really specific and say, 'I am crafting a trading strategy, and these are the kind of things I'm trying to optimize for,'" like whether you're thinking about short or long-term strategy, Ghose says. "So that as [the agent] ingests information, it can treat it with the right context."
CNBC Select is editorially independent and may earn a commission from affiliate partners on links.