AI, Customer, and Enterprise
Most bank AI bans are really policy statements, not behavior changes. The work just goes underground.
In retail and commercial banking ops, people still have to write customer emails, summarize regulations, draft credit notes, and turn meeting chatter into decisions. When LLM use is prohibited, the pressure does not disappear - it shifts into personal accounts, copy-pasted data, and zero audit trail. That is a risk management and compliance problem, and it shows up fast in an audit.
BBVA (BBVA) went the other way: ChatGPT (OpenAI) Enterprise access for 120,000+ employees after a pilot reported about 3 hours saved per person per week on routine work. The part that matters is not clever prompting. It is vendor due diligence, data privacy guardrails, clear do and do not rules, and a few approved workflows where time savings are real and trackable.
This makes me wonder how many 2026 cost plans are still counting the ban as “control,” while the actual cost sits in shadow AI risk and lost hours.