1 week ago
18
Companies are handing AI agents more independence at work. They have not yet decided who answers for it when those agents get something wrong.
That’s the core finding of a new report from the Wharton School of the University of Pennsylvania, produced with research firm GBK Collective. When asked who should be accountable after an AI system fails, respondents did not converge on an answer. They split into near-equal camps, like a jury that can’t even agree which defendant is on trial.
A three-way tie with no winner
The survey covered 1,416 US enterprise leaders and employees. Researchers asked a simple question: when AI makes a mistake on the job, whose problem is it?
The answers landed in roughly even thirds. The employee who deployed the AI, that employee’s manager, and the company’s risk team each drew approximately 25% of responses.
Around 10% of respondents pointed at the AI vendor. Roughly a similar share said they simply didn’t know who was responsible.
The picture didn’t get much clearer among managers specifically. Within that group, 28% said governance or risk management teams should own an AI failure. Another 26% named the deploying employee’s manager, and 25% placed the blame on the employee.
Then there’s the most telling figure. One in five managers said they were unsure who should be accountable at all.
Agents are moving faster than the org chart
Organizations are increasingly rolling out autonomous AI agents that can carry out tasks independently over extended periods, not just answer a prompt and wait for the next one.
Major technology firms, including OpenAI, Google, and Meta, continue to push AI capabilities designed for office use. According to the report’s findings, many executives are wrestling with the accountability question in real time, and a significant number have not yet thought through what happens when AI makes mistakes.
The study, released on October 13, 2026, is part of the broader Wharton Human-AI Research (WHAIR) initiative.
The ROI story is rosier, sort of
The report also looked at whether AI is paying off. On the surface, the answer is encouraging: 80% of leaders say AI is generating positive returns on investment.
Look one layer down, though, and the enthusiasm thins out. Only 28% report distinctly positive outcomes.
One detail stands out for anyone trying to close that gap. Departments that measure AI against growth outcomes are nearly three times as likely to report strong ROI.
What this means
For companies, the accountability split is less a philosophical puzzle than an operational risk. When responsibility is spread evenly across three groups, it can effectively belong to none of them.
Only around 10% of respondents put responsibility on the AI provider. That indicates most organizations view AI failures as an internal matter, which places the burden of building oversight squarely on the companies doing the deploying.
For employees, the takeaway is more personal. If a quarter of respondents believe the person who deployed the AI should carry the blame, workers using these tools have good reason to ask their employers for written policies before an incident rather than after one.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.
English (US) ·