AI Use Policy
Draxl uses AI to prepare first drafts of high-volume legal-operations work. This policy explains how, and the controls around it, so your practice can assess it against its own professional obligations, including guidance such as ABA Formal Opinion 512.
Drafts only; attorneys decide
Every AI-assisted output is labelled as a draft and enters a review queue. Nothing is finalized, stamped, redacted, produced or sent until an attorney at your practice approves it, and the approval and the approved version's fingerprint are recorded in the audit log. Draxl doesn't give legal advice or make legal judgments.
Checked against the source
Page and line citations, names and documents in an AI draft must exist in the source material. If they don't, or the model fails, Draxl discards the model's answer and uses its built-in rules engine. The model can add findings to the rules engine's results but can't remove them.
Where your data goes
AI drafting runs on Anthropic's Claude models through Amazon Bedrock in the United States. Amazon Bedrock doesn't store prompts or outputs, doesn't share them with the model provider, and doesn't use them to train models. Draxl sends client content only to AI services that keep no copy of prompts or outputs; where none is available, client content is processed by Draxl's built-in engine only. Prompts and outputs aren't kept in usage logs, and Draxl never uses your data to train AI models.
Your controls
A practice admin can switch AI drafting on or off at any time and see AI usage and cost per module. Plans include a monthly allowance of AI drafts; beyond it, Draxl continues on the rules engine.
Your responsibilities
Lawyers remain responsible for supervising the tools they use, reviewing outputs, protecting client confidences, and deciding whether to inform clients or obtain consent about AI use under the rules that apply to them. Questions: mkihiro@immanagement.co.