AI Designed to Prevent Downtime Is Causing New Outages, Splunk Says: why it matters for teams

A Splunk study with Oxford Economics surveying 2,000 Global 2000 executives finds AI systems deployed to reduce outages are introducing new failure modes and driving higher costs for unplanned downtime, which Splunk estimates at roughly $600 billion a year.

A Splunk study, produced with Oxford Economics and based on a survey of 2,000 Global 2000 executives, finds that AI systems intended to prevent operational interruptions are themselves causing outages and increasing financial exposure. Splunk estimates unplanned downtime now costs businesses about $600 billion annually, a roughly 50% rise over two years, underscoring why these failures matter for large enterprises.

The report shows firms are investing heavily to avoid interruptions — a median annual spend of $24.5 million on AI aimed at preventing downtime — yet new risks persist. Half of respondents reported outages tied to incorrect AI automation or model drift, and nearly one‑third traced failures to bugs introduced when AI was embedded in production code. Splunk, a unit of Cisco, frames these outcomes as evidence that current practices are not containing the new failure modes created by production AI.

As deployments shift from copilots to more autonomous agents, organizations are seeing different patterns of failure tied to reduced human oversight. Splunk calls this the “reliability paradox”: the more aggressively AI is used to lower operational risk, the more it can create a less predictable category of risk. Kamal Hathi, Splunk’s senior vice president and general manager, warns many firms deploy AI into mission‑critical systems without defined escalation paths or monitoring tuned to detect model drift.

The financial fallout extends beyond IT line items. Splunk reports every minute of downtime costs roughly $15,000; firms lose an average of $300 million before formally declaring a crisis; stock prices fall about 3.4% after major incidents; ransomware payouts have climbed toward $40 million; and regulatory fines average about $51 million. Taken together, these findings indicate boardrooms, IT and risk teams at Global 2000 firms face growing operational and financial exposure as AI moves deeper into core systems.