Security and policies

AI Policy

Atom uses AI to help researchers and research offices find funding, check eligibility, and prepare proposals. This page explains how we use AI, what happens to your data, and how people stay in control of the results.

How we use AI

  • Finding funding: matching researchers to grants, summarizing opportunities, and explaining why a grant fits.
  • Checking eligibility: reading funder rules and comparing them with a researcher's profile.
  • Preparing proposals: drafting guides, outlines, and budget justifications, and reviewing drafts against the solicitation.
  • Researcher profiles: drafting profile summaries from public information and what researchers share with us.
  • Atom AI chat: answering questions about grants, proposals, and profiles.

Your data

  • We never use customer data to train AI models.
  • Each request includes only the information that task needs. We never send passwords, credentials, or API keys to an AI provider.
  • Requests to AI providers are sent from our servers, never from your browser, and AI models never get direct access to our databases.
  • We log each AI request's feature, model, and usage, but not its content, so we can track cost, errors, and reliability.

How AI providers handle your data

  • We use AI models from OpenAI, Anthropic, Google, Perplexity, and TypeSafe.
  • We use their paid business APIs, never consumer apps or free tiers. Under these business terms, providers do not use the data we send to train their models.
  • We have zero data retention agreements with these providers, so they don't store the requests we send or the responses they return.
  • Customer data only goes to providers whose terms prohibit training on it.

People stay in control

  • AI results are suggestions and drafts for people to review, not decisions.
  • Atom AI acts only within your own account permissions. When it changes an existing document, the edits arrive as suggestions you accept or reject.

Fairness

Gender, race, ethnicity, age, disability, and veteran status never raise or lower how well we rate a grant's fit. Some funders limit who can apply (for example, programs for early-career or underrepresented researchers). We check those rules against the researcher's Atom profile, which we draft from public sources and what the researcher shares with us, and which they can review and edit. When a rule depends on something the profile doesn't say, such as gender, we mark it for the researcher to confirm.

How we manage AI risk

We manage AI risk using the NIST AI Risk Management Framework, reviewing AI features for privacy, accuracy, bias, and security. Before we change the model behind a feature, we compare the options on real examples from that feature. Customers can request our full AI Risk Management Policy.

Questions or concerns

If you see AI output that looks wrong, biased, or concerning, or you have questions about how we use AI, email support@atomgrants.com.

Last updated: November 18, 2026.