Responsible AI & Automation Policy
Human control, evidence grounding, and prohibited automation.
1. Core commitments
GrantPilot separates verified facts from proposed, assumed, missing, and expired information; cites authorized evidence; treats imports as untrusted; exposes warnings and uncertainty; preserves versions; and requires human approval for consequential actions.
2. Prohibited uses
No fabricated evidence, deceptive impersonation, discriminatory profiling, rights-affecting decisions without lawful human review, credential disclosure, access-control bypass, CAPTCHA/MFA circumvention, prohibited scraping, mass applications, spam, or autonomous legal/financial commitments.
3. User responsibility
Review every external output, verify dates and amounts against official sources, confirm attachments and recipients, correct false associations, protect confidential data, and escalate legal, tax, accounting, compliance, eligibility, or contractual questions to qualified professionals.
4. Limitations, reporting, and correction
AI and deterministic scoring can produce errors, omissions, stale information, or inappropriate recommendations. The service displays available sources, confidence, gaps, and review controls but cannot eliminate these risks. Report unsafe or materially incorrect behavior to support@ironhand.org; preserve relevant record references and do not email credentials or confidential evidence.