Version 1.0 · Effective 2026-07-11

How BloomOS uses AI

The question every nonprofit asks first: does an AI train on my donor data?

No. Here's the whole picture, including the parts most vendors leave out.

Who processes it

BloomOS sends data to Anthropic's Claude API and nowhere else. We don't use OpenAI, we don't use Google's models, and we don't run our own. One AI vendor, named, so you know exactly who's in the chain.

Anthropic is listed on our subprocessors page. They hold SOC 2 Type I and Type II, ISO 27001:2022, and ISO 42001.

Does it train on your data?

No. BloomOS accesses Claude through the Anthropic API with a commercial API key, which places our usage under Anthropic's Commercial Terms. Under those terms, Anthropic does not use inputs or outputs from commercial products to train its models by default.

There is one way that could change: an organization can explicitly opt in to Anthropic's Development Partner Program to contribute data for training. We have not opted in, and we won't. If that ever changed, it would be a material change to this page and you'd get notice before it took effect.

How long is it kept?

Anthropic deletes API inputs and outputs from their backend within 30 days of receipt or generation. Conversation content is not retained by default for the API.

Two exceptions you should know about, because they're real:

1. If content gets flagged by Anthropic's automated safety systems as a possible usage policy violation, Anthropic may retain it for up to 2 years. This is their abuse-prevention mechanism and it applies to every customer of theirs, including us.
2. Anthropic offers a stricter zero data retention arrangement, negotiated per-organization through their sales team. We do not have one today. If your organization needs it, tell us, and we'll pursue it.

What actually gets sent

AI features in BloomOS are deliberate, not ambient. Nothing is sent to Claude in the background while you work. Data leaves BloomOS only when you take an action that requires it.

Funder research and prospect briefs. When you run research on a prospect or a foundation, BloomOS sends the name of the organization or funder, the public context it needs, and the relevant fields from your prospect record. Claude searches the public web and returns a brief. What goes out is what's needed to do the research.

Drafting. When you ask BloomOS to draft an email, a grant narrative, or a summary, the content you're drafting from goes with the request. If that content includes a donor's name and giving history, that goes too, because it's what the draft is built from.

The daily briefing. BloomOS generates one written narrative per day summarizing your organization's position. It sends the aggregate numbers and status of your objectives, not your donor list.

Next best actions. Suggestions about which relationships to advance are computed from your pipeline data. Where a suggestion needs written reasoning, the relevant records go with it.

What is never sent: passwords, API keys, payment card numbers, or your authentication tokens. BloomOS doesn't store payment card numbers at all.

Your right to say no

If you don't want any of your data going to an AI vendor, don't buy the tier that includes it. Reed and the AI features live in Bloom Grow and above. The base Bloom tier is a plain operating system: your data stays in the database and goes to Claude never.

If you're already on a tier with AI and you want it off, email us and we'll disable it for your organization.

The judgment call we're asking you to make

AI-written funder research is a draft, not a fact. Claude searches the public web and it can be wrong, out of date, or confidently mistaken about a foundation's priorities. BloomOS labels AI-generated content as AI-generated, and it's on you to check it before you act on it. We designed the product so an AI never sends an email, never records a gift, and never changes a number without a human pressing the button.

That's a deliberate limit, and we'd rather explain it than let you assume the robot is smarter than it is.

Questions: hello@bloomos.org.