United Airlines Blocks AI Agents from Booking Flights

United Airlines is taking a hardline stance on artificial intelligence agents by restricting them from booking flights through its website, third-party travel agencies, and flight aggregators. The decision reflects growing concerns among major corporations about the security and privacy implications of letting autonomous AI systems handle sensitive transactions.

The airline’s move comes as more companies wake up to the risks posed by agentic commerce. JPMorgan Chase and Amazon have already implemented similar restrictions, recognizing that handing over payment details and login credentials to AI agents opens the door to potential disasters.

Why the Risk?

United’s concerns are not hypothetical. Allowing an AI agent to complete a booking could expose users to “unintended purchases, account changes or use of MileagePlus miles,” according to the airline. Beyond financial exposure, there are regulatory headaches. The Department of Transportation and Federal Aviation Administration have existing rules that create friction with AI automation. For example, only a human can legally acknowledge that their bag doesn’t contain explosives during flight check-in. That simple requirement alone disqualifies AI agents from handling the complete booking process.

There’s also the embarrassment factor. Meta’s Muse AI application famously leaked a user’s address online, a cautionary tale that has made executives nervous about associating their brands with tech failures. United CEO Scott Kirby has been refreshingly honest about this concern, telling attendees at a recent event: “To be honest, I don’t want to create another middleman.” His reluctance reflects a broader anxiety about AI becoming yet another layer between companies and their customers.

Not every airline agrees with this approach. Southwest Airlines, notably, created a plugin allowing users to book flights directly through ChatGPT. That contrarian move suggests the industry remains divided on how to navigate the AI agent landscape.

The Reality Check

Here’s the encouraging news for skeptics: agentic commerce appears to be struggling to gain real traction with consumers. Uber’s recent earnings call revealed telling numbers. During Halloween week, the company handled 320 million trips and deliveries. Of those, AI agents were responsible for a measly 500 Uber Eats orders and only four rides. These figures suggest that despite the hype, most people aren’t ready to hand their digital lives over to autonomous bots.

The distinction matters here. Handing an AI your credit card number is fundamentally different from sharing your email login, which is itself different from asking a bot to send you a daily news summary. Each scenario carries different risk profiles and different levels of user friction. It seems like letting an agent shop for you isn’t as seamless or reliable as it needs to be to achieve mainstream adoption.

The proliferation of AI apps like Muse might suggest otherwise, but downloading an app is not the same as trusting it with your money and sensitive personal information. That final step of actually delegating transactions appears to be where agentic commerce hits a wall.

What This Means

United’s decision to block AI agents signals that major corporations are choosing caution over innovation hype. The combination of security vulnerabilities, regulatory complications, and genuine user hesitation creates a moment where the technology’s limitations are becoming impossible to ignore. Whether this represents a temporary pause or a fundamental limitation of current agentic systems remains to be seen.

What’s clear is that the frictionless future some promised for AI commerce isn’t arriving on schedule. And for tech companies betting on autonomous agents, that gap between promise and reality might be harder to bridge than expected.

Source: Wall Street Journal

If the most powerful technology companies are still too nervous to trust AI agents with transactions, what does that tell us about the actual state of artificial intelligence maturity?