- 53% of British CIOs have seen at least one AI agent violate a policy and impact either the business or a customer
- The global average for rogue AI actions is just 31%, raising questions on AI governance in the UK
- CIOs are under increasing pressure to deliver value from AI investments, with a majority expecting budgets cuts and even unemployment if targets are not met
AI agents going rogue is a growing challenge for British businesses according to a new study which also raises questions over policies and governance of AI in the workplace. Rule breaking AI agents in the UK are apparently experiencing far greater problems than the global average.
The research, conducted by Harris Poll on behalf of Dataiku, reveals that British-based CIOs are also under immense pressure to deliver on AI investments.
CIOs blame what Dataiku calls “AI-Agent sprawl” – the creation and use of custom AI agents in the workplace, which are deployed faster than they can be governed. This practice can potentially open businesses to risks, not just for the organization itself, but also for customers.
AI agent risks
Harris Poll’s survey of 685 CIOs in the UK, France, Germany, US, UAE, Japan, South Korea, and Singapore reveals governance challenges facing CIOs (Chief Information Officer) around the world. But the results show us that the UK seems to be facing some unique issues.
The statistic that 53% of British CIOs have revealed that they have seen one or more AI agents violating a business policy that has impacted their own organization or a customer is noticeably higher than the average of 31%. But the problems don’t end there. Only 49% could produce an audit trail to explain the actions of an AI, and 68% believe it would take over 24 hours to identify a rogue AI agent and limit its impact.
With the challenge of AI sprawl cited by 87% of British CIOs, the problems seem to derive from the policies, procedures, and laws governing AI use.
“More than half of UK businesses reporting that an AI agent has violated policy should be a signal that monitoring alone is not enough," noted Faye Murray, Field Chief Data Officer at Dataiku.
The impact of a rogue action needs full accountability, something that Murray observes “...cannot be retrofitted after an incident; they need to be designed into the deployment from the start.”
Unplugged
The implication is that the industry has got things wrong when it comes to AI. Autonomous systems can’t be simply left to get on with things, at any scale. Dataiku’s CEO Florian Douetteau says: “The industry sold CIOs on autonomy and forgot to mention the cost: autonomous systems need to be managed, not just monitored.”
"Monitoring tells you an agent is running. Managing tells you whether it's earned the right to keep running. Right now, almost nobody can fire an agent.” For those businesses where an AI agent has got out of hand, that may make for uncomfortable or disappointing reading, but the solution really is improved oversight.
As Douetteay says, the only alternative is to “unplug it and hope."
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
English (US) ·