Real bank data at runtime: Azure OpenAI only

← Tips · Compliance & safety

Real bank data at runtime: Azure OpenAI only

If a tool will process real bank data at runtime, its LLM calls MUST go through Azure OpenAI. The public Anthropic API and Gemini are off-limits for real customer data. This is data-residency, not a preference.

advanced

The team’s hard rule: runtime LLM calls that touch real bank data go through Azure OpenAI. Period.

Why: data residency. The Anthropic API and Google Gemini do not give us the data-residency and compliance guarantees the bank requires for real customer data. Azure OpenAI does.

Which API for which job:

• Anthropic API (Claude) — best reasoning quality. Use freely for non-confidential workloads (research, internal docs, public data).
• Azure OpenAI — mandatory for real bank data at runtime. Always.
• Gemini — large-context tasks (huge documents). Non-confidential only.

This applies to runtime LLM calls (the tool calls an LLM each time it runs). It does NOT apply to using Claude Code itself for build-time work — Claude Code is the AI assistant we use to write the tool, not the LLM the tool calls in production.

When in doubt, ask. Compliance review beats data-residency incident.

Next tip →

What a real CLAUDE.md looks like (worked example)

A worked example of a project-level CLAUDE.md, distilled from Anthropic's official guidance, Karpathy's viral behavioral rules, HumanLayer's "under 60 lines" guide, and the awesome-claude-md curated index. Each section is annotated with why it earns its slot.

Open it → Back to all tips