Taco Bell AI roll-out in disarray after customer orders 18,000 cups of water
Taco Bell AI roll-out in disarray after customer orders 18,000 cups of water
Occurred: August 2025
Page published: September 2025
A prank order for 18,000 cups of water crashed Taco Bell’s new AI-powered drive-through order system, exposing serious flaws in its automation governance and technology.
Taco Bell's new voice AI ordering system malfunctioned dramatically when a customer exploited its lack of safeguards by ordering 18,000 cups of water in one transaction, forcing a crash and drawing ridicule.
Introduced at over 500 drive-through locations since 2023, the system is meant to streamline operations and enhance the experience of customers.
But the system proved highly unreliable. Other misshaps included it insisting customers add extra drinks, refusing substitutions (such as replacing meat with beans), misinterpreting items, leading to wrong orders like extra chalupas, telling customers restaurants were out of everything except drinks and sauce packets, and confusing customers.
Taco Bell's AI drive-through incident stemmed from a mixture of technical, operational, and governance failures, likely amplified by internal pressure to deploy AI quickly without fully considering how custmers might use and abuse the technology.
The system did not have adequate controls to flag or block unreasonable orders, such as the request for 18,000 cups of water, exposing it unduly to exploits and pranks.
In addition, there was now human-driven review designed into the system to identify or halt obviously unrealistic transactions, and Taco Bell's governance model did not empower staff to intervene quickly on an ad hoc basis in cases where the system was failing.
The incident highlighted major vulnerabilities in Taco Bell's AI ordering system and raised questions about the company's governance and technology strategy and practices.
Omilia Voice AI Solution🔗
Developer: Omilia
Country: USA
Sector: Travel/hospitality
Purpose: Process customer orders
Technology: Speech-to-text; Voice recognition
Issue: Accuracy/reliability
Incident no: AIAAIC2019