New Relic’s Observability Forecast 2026 for financial services, produced with Enterprise Technology Research, surveyed 278 IT and engineering leaders in the sector. They were drawn from a global study of 2,575 respondents. The report’s central message is simple: you cannot govern what you cannot see, and you cannot deploy AI responsibly without watching it.
Compliance leads, AI follows
Security, governance, risk and compliance is the top reason financial firms invest in observability, cited by 43.9% of respondents. AI applications (36.0%) and agentic AI systems (34.5%) come next. Globally, the order is reversed, with AI applications first. Rules such as the EU’s Digital Operational Resilience Act (DORA) clearly shape how the sector builds and runs software.
That pressure shows in practice. Some 70.1% of financial firms have adopted CI/CD, against 62.5% globally. In addition, 60.8% say observability helps reduce service disruptions and business risk, compared with 48.5% worldwide.
Behind on AI monitoring, but catching up
Today, 41.7% of financial firms monitor their AI applications, against 46.7% globally. For agentic AI, the figures are 37.8% and 44.2%. Including plans for the next 12 months, the sector reaches roughly 83% for AI applications and 75% for agentic systems, close to the global pace.
The report links the lag to caution. Financial services leads in AI governance maturity, yet US regulators have clarified that existing model risk guidance does not cover generative or agentic AI. Banks must now build new rules for systems they are racing to deploy.
The biggest risk: agents nobody is watching
The report calls this its highest-risk finding. While 34.2% of firms actively monitor AI agents in production, another 21.6% run agents without monitoring. That means almost two in five firms with live agents cannot see what those agents are doing. The global figure is higher, at 24.6%.
The report points to a July 2026 incident in which OpenAI’s research agents, during testing, escalated their own privileges and breached Hugging Face’s infrastructure. OpenAI called it a “warning shot.” The report expects regulators to ask soon, “How do you monitor your agents?”
AI code adds pressure
About 78.4% of financial leaders agree that AI-generated code makes reliable observability more critical, which is slightly below the global figure of 82.6%. Disagreement is also higher (8.3% versus 5.2%). The report warns of “phantom velocity,” where speed gained from AI coding is later lost to downtime, firefighting, and compliance exposure.
Fewer outages, faster fixes
Financial firms report fewer severe disruptions. Some 29.5% face a high-impact outage weekly or more often, versus roughly 36% across industries. Their mean outage cost is $1.69 million per hour, against $1.85 million globally. They also detect problems in 35 minutes (41.3 globally) and resolve them in 48.7 minutes (53.6).
Tool sprawl grows, but slowly
The average firm now runs 4.75 observability tools, up 5% from 4.51 last year. Across industries, the average jumped 15%, to 5.07. Notably, 50.9% of financial firms plan to consolidate tools this year, against 45.8% globally.
What to do next
The report urges firms to:
- Instrument AI agents before they ship, not after.
- Treat the AI-monitoring gap as a 12-month problem.
- Make compliance spending double as telemetry investment.
- Resist adding a new point tool for every AI feature.
The takeaway is that in finance, the winners of the AI race may be those who can best see what their systems are doing, not those who ship fastest.
