Businesses are handing AI more power, but there is no clear answer on who takes the blame when it goes wrong. A new Wharton School study at the University of Pennsylvania finds managers deeply divided on AI accountability when systems make consequential errors.
The survey covered more than 1,000 business leaders, including managers, directors, executives, owners and founders. Roughly a quarter held the employee using the AI tool responsible, while another quarter pointed to that employee’s manager. A similar share named the company’s risk team. About 10% blamed the AI vendor, and about the same number were unsure.
Small Prompt Errors, Big Consequences
The most immediate risk may not be rogue AI but simple mistakes. Leeron Walter, a vice president at Teramind, told Bloomberg that a vague instruction can snowball once an agent can reach live tools such as email or customer data. He cited a major US retailer where a flawed instruction led an AI bot to send false refund and cancellation messages to over 18,000 customers in under six hours. In another case, a consumer electronics company’s AI-posted pricing listed a flagship device at 85% off for three hours. More than 700 orders came in, and customer service spent weeks dealing with the fallout.
AI Agents Are Gaining More Autonomy
The question is growing more urgent as tech giants release more capable agents. OpenAI recently launched Dots, always-on agents that connect to over 4,000 apps and keep working between check-ins. Google announced a Gemini enterprise agent that can plan and complete multistep work across company systems. Meta’s Muse personal agent handles email and travel bookings and passed 2.5 million downloads in its first two weeks, according to Sensor Tower data.
Consumers are adopting AI too. IBM research shows 41% use AI assistants to research products and 31% use them to find deals. Without clear ownership of AI decisions, the Wharton findings suggest, companies may struggle to respond when the next serious error hits.