Using the AI responsibly.
Barkrow applies AI to reconcile insurance documents. This policy sets out how the AI may and may not be used, and how it handles your documents.
It supplements the Terms of Service and the general Acceptable Use Policy, and applies to everyone who accesses the platform.
A tool for a licensed professional
Barkrow is decision support for a licensed broker, not a replacement for one, and not legal, coverage, or insurance advice. Do not rely on an output as a coverage opinion or a statement of what a policy covers. The broker remains responsible for the bind and for every coverage decision.
Keep a human in the loop
Every extracted value carries a citation to its source clause and page. Review the source before relying on an output, and review it again before you send it to a client, carrier, or anyone else; you remain responsible for what you send. Confidence is guidance about where to look first, not a guarantee of correctness. Where Barkrow cannot cite a value, it says so rather than guessing.
Permitted use
Use the platform on documents you are authorized to handle, for legitimate commercial-insurance compliance work, within the seats and scope of your agreement.
Prohibited use
Do not attempt to defeat or reverse the PII boundary, or to recover raw identifiers from tokenized data. Do not submit documents or data you lack the right to process. Do not reverse-engineer, scrape, or resell access; do not use outputs to make automated decisions without human review, including decisions about a person’s eligibility for insurance, credit, employment, or housing; and do not use the platform for any unlawful, discriminatory, or deceptive purpose. Do not attempt to extract the underlying model, prompts, or training data, to manipulate the AI into ignoring its instructions, or to use the service to build a competing product.
Data you may submit
Submit only documents you are authorized to handle, redacted where your obligations require. Do not upload special-category personal data, protected health information, payment card numbers, or government identification numbers beyond what a coverage review genuinely needs.
How the AI handles your documents
Before document text reaches a model, Barkrow replaces detected sensitive identifiers with tokens. Automated detection has limits, and some names or other identifying information may remain. Language-model inference runs on Anthropic’s platform, called directly, with a United States processing preference. Text embeddings for search are computed on Cloudflare’s network. Both providers are listed on the Subprocessors page.
As of 2026-10-11, Anthropic’s organization setting shows that API data is retained for 30 days by default. Zero data retention is not currently verified. This setting does not establish whether particular customer content was processed or retained. Existing data-processing commitments remain applicable; this factual update does not waive or replace them.
No training on your content. We claim no right to train models on your documents, and no setting turns it on. See the Data Processing Addendum.
Model limitations
AI can be wrong or incomplete. Outputs may miss an endorsement or misread a form edition, and similar documents can produce similar outputs for different customers. Verify against the cited source. We do not warrant correctness beyond what the agreement states.
Enforcement
Misuse may lead to suspension of access. We cooperate with lawful requests and reserve the right to act on violations of this policy or the Terms of Service. Report misuse or an output that looks unsafe to security@barkrowsystems.com.
Questions? Write to legal@barkrowsystems.com.