Version 2026-09-06 · Effective 6 September 2026
1. Two separate modes
Local structured mode runs completeness rules and drafting logic in the browser without sending workspace content to RCET or an AI provider. Optional AI enhancement is a separate user-initiated action. It sends only the selected draft after an explicit classification confirmation.
2. Current permitted data
3. No silent or automatic AI transmission
RCET will not send an entire workspace merely because a user opens, types in or saves the local tool. AI processing occurs only when the user selects “Enhance with AI”, confirms the material is training or fictitious, and submits the request. The returned version is shown separately and does not overwrite the user’s draft without a further confirmation.
4. What the AI may do
The approved tasks are improving clarity and structure, preserving chronology, identifying ambiguity, organising supplied material, and producing human-review prompts. The model is instructed not to invent facts and to mark uncertainty.
5. Prohibited reliance and decisions
AI must not determine credibility, truthfulness, guilt, innocence, reasonable suspicion, probable cause, arrest, charge, prosecution, sentence, use of force, risk level, employment action, legal rights or any decision producing a legal or material effect on a person. It must not be treated as legal advice, evidence, an expert opinion or an authoritative statement of law or policy.
6. Human review duties
The user must compare output with source material, confirm every fact and quotation, verify dates and calculations, check applicable law and policy from authoritative sources, remove unsupported statements, preserve exculpatory and adverse material, record uncertainty, obtain required supervision and ensure the final document reflects the responsible human author’s judgement.
7. Model limitations
Generative AI is probabilistic and may be inaccurate, incomplete, inconsistent, biased or overly confident. Similar inputs may produce different outputs. The model may not know current law, policy, local practice or the full context. A fluent answer is not proof of correctness.
8. Provider and data handling
When enabled, RCET uses a business API configuration with response storage disabled in the request. RCET does not use workspace content to train its own model and does not save AI prompts or outputs to the RCET database. The external provider’s contractual processing, security, retention and subprocessors must be reviewed and approved before any production-data use. The local mode remains available if AI is disabled or unavailable.
9. Production approval gate
Use with live or sensitive information requires a documented data-protection impact assessment, lawful basis, retention and deletion rules, international-transfer safeguards, provider agreement, access controls, incident response, security testing, organisational authority, user training and a written change to the current restriction.
10. Reporting concerns
Do not continue using output that appears harmful, discriminatory, fabricated or unsafe. Preserve only the minimum non-sensitive information needed to report the issue and contact contact@cuttingedgetools.xyz.