United States – Around 22% of financial services companies have deployed artificial intelligence (AI) agents in production without monitoring them, raising concerns over compliance, operational risks and visibility into automated systems.
Research from New Relic, based on responses from 278 financial services IT and engineering leaders and practitioners within a global study of 2,575 respondents, found that regulatory compliance was the leading driver of demand for observability tools in the sector. AI applications ranked second, followed closely by agentic AI systems.
The findings highlight a gap in AI oversight, with financial services firms lagging behind the broader market in monitoring AI applications and agentic systems. Some 42% of financial services companies monitor AI applications, compared with 47% globally, while 38% monitor agentic systems, against 44% across industries.
However, adoption is expected to increase over the next 12 months. The proportion of financial services companies monitoring AI applications is projected to reach 83%, exceeding the global projection of 80%. Monitoring of agentic AI systems is expected to rise to 75%, approaching the global forecast of 77%.
The research also found that 78% of financial services leaders agreed that AI-generated code makes reliable observability more critical.
Despite the monitoring gap, the sector recorded stronger operational performance than the cross-industry average. Around 30% of financial services organisations experience high-impact IT outages weekly or more often, compared with 36% globally.
When outages occur, financial services companies take an average of 35 minutes to detect them and 49 minutes to resolve them. The corresponding global averages are 41 minutes and 54 minutes, respectively.
The average cost of a high-impact IT outage in financial services was US$1.69 million per hour, below the global average of US$ 1.85 million. The sector also reported higher adoption of certain automation practices, with 70% using continuous integration and continuous delivery (CI/CD), and 51% using automated incident response.
The findings come amid growing regulatory demands on financial institutions, including the European Union’s Digital Operational Resilience Act, which entered into application in January 2025. Monitoring AI agents is particularly important where automated systems can move money or alter configurations, as unobserved actions may create operational and compliance risks.
“In a highly regulated industry like financial services, innovation cannot outpace governance,” said Nic Benders, Chief Technical Strategist at New Relic.
“The data shows that while financial institutions are taking a cautious approach to deploying AI-native tooling, the race is on. To maintain their strong operational discipline and avoid unseen compliance failures, these organisations must close the visibility gap and instrument their AI agents before they ship. You simply cannot deploy AI responsibly without observing it,” Benders continued.
