Product guidance

Responsible AI use

Use Cannon to accelerate review while keeping provider awareness, source verification, professional judgment, and final approval with the legal team.

Updated
August 11, 2026
Version
2026-08-11.1
Product
Cannon Caseworks

Assistive, not autonomous

Cannon organizes records, identifies support and gaps, answers bounded questions, and prepares drafts. It does not replace a lawyer's judgment, duties to the client and tribunal, or responsibility for the final work.

What OpenAI receives

A requested generative AI feature may send extracted text, selected excerpts, limited metadata, case-specific review criteria, prompts, questions, images, audio, or bounded case context to OpenAI. Cannon sends this material to return the requested summary, label, category, transcription, description, analysis, answer, draft, or report.

OpenAI

Cannon does not opt API content into OpenAI training and uses non-storage settings where supported. Standard OpenAI API controls can still retain prompts and responses in abuse-monitoring logs for up to thirty days unless enhanced contractual controls apply.

Google Document AI

Google Document AI may receive documents and images for document processing and OCR. It is not Cannon's generative model provider. Do not assume that zero retention, regional processing, HIPAA, or customer-managed-key controls apply unless Cannon confirms them in a signed agreement.

Verify every material claim

Inspect citations and outputs against the underlying record and controlling authority. A citation can be incomplete, ambiguous, or insufficient even when quoted text is accurate. Labels and confidence scores are triage aids, not final legal determinations.

Preserve uncertainty

Treat missing proof, conflicting records, and unanswered intake questions as unresolved. Do not convert an inference into a fact or allow a generated summary to replace the source record.

Review before external use

A qualified lawyer should review analysis, privilege calls, redactions, exports, productions, client communications, filings, discovery material, and work product before relying on or sharing them.

No consequential or biometric decisions

Do not use Cannon as the sole basis for a decision about employment, housing, education, credit, insurance, health care, legal services, government benefits, or another legally significant opportunity. Do not use it to create or match faceprints, voiceprints, or other biometric identifiers, conduct unlawful surveillance, or discriminate unlawfully.

Use authorized inputs

Upload only material the firm is authorized to process through the configured providers. Apply the firm's confidentiality, privilege, legal-hold, protective-order, outside-counsel, client-instruction, and cross-border-transfer requirements.

Report problems

If an answer appears unsupported or a control behaves unexpectedly, stop the affected workflow, preserve the source context, and report the issue to Cannon before external use.