Is it safe to use Claude Code
with financial data?
Short answer: Claude Code runs on your computer but sends your prompts and the contents of files it reads to Anthropic. On Free, Pro and Max plans, data may be used for training if that setting is on (5-year retention; 30 days if off). Team, Enterprise and API data is not used for training by default. Learn on fictional data and minimize real data.
Claude Code runs on your computer, but it works by sending prompts and the file contents it reads to Anthropic's servers. So "it runs locally" does not mean "the data stays local". Whether that is acceptable depends on your plan settings, your employer's policy and the data itself.
What Claude Code sends
To answer you, Claude Code sends your prompts and whatever file contents it reads to Anthropic's API. If it opens your transactions CSV, those rows are sent. It also sends operational telemetry. According to Anthropic's documentation, metrics "never include your code, prompts, or file paths", and error reports have known patterns of secrets, paths and personal information redacted before they leave your machine.
Is my data used to train models?
- Consumer plans (Free, Pro, Max): Anthropic states it will train new models using data from these accounts "when this setting is on", including Claude Code use. You can change the setting at claude.ai/settings/data-privacy-controls.
- Commercial terms (Team, Enterprise, API): Anthropic states it does not train generative models on code or prompts sent to Claude Code under commercial terms, unless the customer chooses to share data for model improvement.
How long is it kept?
- Consumer accounts that allow data use for model improvement: 5-year retention.
- Consumer accounts that do not: 30-day retention.
- Commercial accounts: 30-day standard retention.
- Transcripts you submit through
/feedback,/bugor/share: retained for 5 years. - On your own machine, Claude Code stores session transcripts in plaintext under
~/.claude/projects/for 30 days by default. Transcripts of sessions you started or last continued in Claude Desktop or Cowork are kept at any age unless you setdesktopSessionCleanupPeriodDays.
A practical checklist before using real data
- Learn and build on fictional data. Our sample CSV exists for this reason.
- Check your employer's or client's policy on AI tools. Many firms require an approved commercial plan.
- Review your training setting on consumer plans.
- Minimize the data. Remove account numbers, names and anything the dashboard does not need.
- Separate building from running. Once the app works, you can run it on new files without asking the AI to read them. The AI builds the tool; the tool processes the data.
- Remember local transcripts in
~/.claude/projects/on shared or managed machines. - Turn off optional traffic if required. Anthropic documents environment variables such as
DISABLE_TELEMETRY=1,DISABLE_ERROR_REPORTING=1andCLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC.
Frequently asked questions
Does Claude Code upload my files?
It sends the contents of files it reads, along with your prompts, to Anthropic to generate responses. Files it never opens are not sent.
Does Anthropic train on my Claude Code data?
On Free, Pro and Max plans, only when the model-improvement setting is on. Under commercial terms (Team, Enterprise, API), Anthropic states it does not train on Claude Code prompts or code unless the customer opts in. Checked September 24, 2026.
Can I use Claude Code with client data as a bookkeeper or accountant?
Check your engagement terms, professional obligations and firm policy first. Build and test with fictional data, minimize what the AI reads, and prefer a commercial plan if client data is involved.
Turn a CSV into a working cash flow dashboard.
A self-paced Claude Code workshop for bookkeepers, accountants and FP&A: six lessons, one project, sample data with expected totals. Taught by CFO Leo Godnik.
See the six lessons — $69 → Or start with the free lesson ↗