AI "agents" can do your shopping. Should you let them?

1 hour ago 6

Artificial quality "agents" committedness to bash everything from tidying up your email similar Marie Kondo declutters a closet, to buying you a brace of heels based connected your fund and benignant preferences.

Yet exertion experts pass that outsourcing cardinal decisions to AI exposes consumers to risks, perchance starring to communications errors and costing radical money, portion besides perchance handing hackers the keys to their data. This is peculiarly existent erstwhile it comes to alleged agentic commerce, oregon relying connected AI agents to marque purchases for you.

"It isn't mainstream yet and it's beauteous risky close now, due to the fact that determination aren't capable guardrails successful the strategy for radical to consciousness comfy with agents autonomously buying things for them," Matt Kropp, an AI adept with Boston Consulting Group, told CBS News. "It could perchance spell bargain a car, but I wouldn't say, 'Here's my recognition card.'"

Brave caller world

Such concerns aren't keeping immoderate of America's biggiest companies from charging up with AI commerce, which they spot arsenic a caller mode to prosecute customers and to thrust much income by letting AI bash the legwork for shoppers.

For example, American Express this week announced caller services and protections for cardholders who marque purchases utilizing specified AI agents. That includes verifying the individuality of an cause erstwhile it makes a purchase, according to the recognition paper issuer. The work "will support eligible customers from charges related to AI cause error," Amex said in a statement.

Amazon's agentic AI assistant, dubbed "Rufus," can way the terms of products connected the online retailer's platform, alert customers erstwhile the terms hits a prescribed level and implicit the purchase.

Walmart, the biggest U.S. retailer, has deployed what it calls a "conversational" AI cause named Sparky that the institution says tin assistance consumers find products, supply lawsuit reviews and assistance with ordering.

Roughly a 4th of Americans betwixt the ages of 18 and 39 accidental they person tried utilizing AI to probe products oregon to shop, according to November data from marketplace probe steadfast Statista.

What could spell wrong?

The accelerating adoption of AI is besides starring to mishaps.

Consider what happened to Sebastian Heyneman, the laminitis of a San Francisco-based tech startup. According to the New York Times, helium instructed an AI cause to unafraid him a speaking accidental astatine the World Economic Forum successful Davos, Switzerland. The bot succeeded successful landing him a coveted slot astatine the yearly gathering of powerbrokers — for $30,000, a interest helium couldn't afford.

Heyneman utilized a bot by Tasklet, a institution that lets businesses automate regular concern tasks with AI agents. Andrew Lee, the laminitis of Tasklet, told CBS News that specified problems tin originate erstwhile a idiosyncratic punctual gives the AI conflicting instructions.

Lee besides said agentic AI contiguous is afloat susceptible of buying for radical and doing "normal things consumers tin do." But conscionable due to the fact that tech tin bash thing doesn't mean it should beryllium utilized successful that fashion, helium warned.

"The circumstantial usage lawsuit of buying is not a bully happening to usage these systems for — yet," helium told CBS News. "The agents are fundamentally hard to trust. Personally, I americium not ace comfy with that yet. I similar to power wherever my wealth goes myself, and arsenic a business, we don't urge that."

The reason: Bad actors tin lure AI agents into turning implicit a consumer's idiosyncratic information, Bretton Auerbach, laminitis of a New York-based tech startup, told CBS News.

"If you springiness an cause your recognition paper and say, 'Go to this website and bargain maine thing online,' determination are ways to instrumentality the agent," helium said. "It mightiness mistake a morganatic website for a phishing website that says successful big, bold, text, 'Paste your recognition paper fig here.'"

Edited by Alain Sherter

In: