Over half of UK firms say an AI agent went rogue
Dataiku's global CIO survey found UK firms report far higher rogue AI incidents than the worldwide average
More than half of British Chief Information Officer (CIO) have watched an AI agent break company policy in ways that hit their business or their customers, according to a new Harris Poll survey conducted on behalf of Dataiku in 2026.
The rate of 53% in the UK is much higher compared to the global average of 31%, which implies that there is a unique problem of governance within British organisations.
This survey involved 685 CIOs in eight countries: the UK, France, Germany, the USA, the UAE, Japan, South Korea, and Singapore. Although rogue AI cases happen everywhere, the rates in the UK look more concerning.
Namely, only 49% of CIOs from the UK could provide the audit trail to describe what an AI agent did, while 68% claimed that it would take more than 24 hours to detect a malicious AI.
Dataiku points to what it calls AI-agent sprawl, the rapid creation of custom agents across a business faster than governance structures can keep up. Eighty-seven per cent of UK CIOs cited this sprawl as a direct challenge.
Dataiku Field Chief Data Officer Faye Murray noted that the figures regarding policy violations highlight the fact that monitoring alone is not enough and that there has to be an accountability structure created as part of the deployment process itself and not as an afterthought in case of problems.
Dataiku CEO Florian Douetteau noted that the basic problem lies in communication failures in the industry, with companies being led to believe that an AI-based system is capable of running on its own without informing them of the need to manage such systems.
He pointed out the difference between observing an agent and deciding whether it needs to continue to be run at all, which, he noted, most companies cannot do with their current setup.
