AI downtime costs surge as automation creates new risks

Businesses are spending millions on artificial intelligence to prevent outages and operational failures, yet many are discovering that the technology is creating a new category of risk.

According to a 2026 report by software company Splunk, conducted with Oxford Economics, unplanned downtime now costs businesses an estimated $600 billion annually, a 50% increase over the past two years. Half of all surveyed organisations experienced downtime because of flawed AI automation or model drift.

For many years, corporations thought of AI as the answer to operational uncertainty. It was a simple idea: automate decision-making, avoid mistakes and prevent disruption before the customer even notices anything wrong.

According to the results of a survey among 2,000 corporate executives representing Global 2000 firms, the picture turned out to be quite different. Organisations spend a median of $24.5 million each year to invest in AI systems that should help reduce the risk of downtime. But at the same time, almost a third claimed that outages occurred due to bugs that appeared after AI deployment.

Every minute of downtime now equals about $15,000 for business; the annual losses amount to about $300 million per company.

Contrary to system outages in general, AI-specific outages usually happen gradually. Splunk director of developer evangelism Greg Leffler explained that one of the most prevalent issues is model drift, in which the AI makes decisions by referring to old training data.

AI-specific outages can affect several interrelated services long before any engineer identifies the issue. In addition, faulty integration between different systems remains another issue that involves AI decision-making based on insufficient data, causing disruption to multiple platforms.

However, the survey revealed that only 38% of tech executives are able to identify the reasons behind such incidents on a consistent basis.

While 44% of organisations already use agentic AI systems, 68% worry that these tools may behave unpredictably.

Nearly one in four organisations reported encountering prompt injection or data poisoning attacks designed to manipulate AI behaviour. Meanwhile, 77% of technology leaders believe cybercriminals using generative AI will increase downtime risks in the future.