On May 7, the OCC released its spring 2026 semiannual Risk Perspective report, identifying key risks facing the federal banking system. The OCC reported that bank earnings improved in 2025, supported by loan growth and a decline in funding costs, with balance sheets remaining strong and capital ratios and liquidity high by historical standards. The report highlighted that aggregate system-wide credit risk remains manageable but flagged credit conditions and refinancing risk in certain segments of commercial real estate and private credit markets as warranting ongoing monitoring. The agency observed modest increases in past-due loans in some consumer portfolios, though OCC-supervised banks’ exposure to borrowers with weaker credit scores remained manageable.

The OCC also highlighted operational and compliance risks, noting that AI has reshaped the cybersecurity environment by making it easier for bad actors to launch attacks and enabling faster, more complex intrusions targeting financial institutions, while also enabling new forms of fraud. At the same time, the report noted that banks choosing to incorporate more stringent security measures, such as multifactor authentication and timely patch management, and use AI-developed tools to support cybersecurity operations and vulnerability monitoring, will be better positioned to mitigate AI-enabled cyber risks. The report further warns that state-backed cyber actors and organized cybercriminal networks remain a persistent threat, with risk heightened by geopolitical tensions tied to the Middle East conflict, which the report states may strain bank compliance systems and increase the potential for sanctions and BSA/AML violations.

The report references the OCC, Fed, and FDIC’s recent overhaul of their model risk management framework — which, in part, explicitly excluded generative and agentic AI from its scope (covered by this Orrick Insight here) — noting that the agencies plan to issue a request for information on banks’ use of AI, including generative and agentic models, which may inform future supervisory expectations.