Updates
Governance compliance AI is on by default (GR) The AI Compliance Advisor — gap analysis, audit preparation, AI suggestions, and the AI prompts workspace — is now enabled for every organization on first load. Previously the master toggle defaulted to OFF and admins had to flip it before any GR AI page would render. Existing organizations have been migrated; new organizations start with the advisor on. The off-switch still lives at GR → Settings → AI → Enable AI Compliance Advisor; flipping it off pauses every GR AI surface without deleting prior suggestions. See the AI Compliance Advisor admin guide. Clinical AI documentation is on by default with published DSI disclosures (CL) AI-assisted progress-note drafting (CL-65) is now enabled for every existing organization. The platform has also seeded published Decision Support Intervention (DSI) disclosures for the two PHI-lane models the feature uses —anthropic/claude-sonnet-4.6 and openai/gpt-4o — so the §170.315(b)(11) interlock is satisfied on day one. The clinician still reviews, edits, and signs every draft. Billing facts (service date, times, duration, diagnosis) remain locked to the encounter and cannot be introduced by the model. The per-org off-switch lives in CL → Settings → AI Documentation. Organizations created after this release start dark — an org admin publishes the required DSI disclosures in the CL-65 UI before AI drafting unlocks (fail-closed by design). See the AI Documentation card in Clinical settings.
Offer letter wizard reachable for every org (HR)
The FEATURE_HR_JOB_OFFER_WIZARDS platform flag is now seeded enabled, so the guided Offer Letter Wizard launches from both /hr/ats/applications/:id/offer and /hr/ats/offers/new without a platform admin first flipping the flag. The hr.ats.offers.create permission still gates the route. Admins who want the wizard off can disable the flag through the platform feature-flag admin.
Bug fixes
- AI structured-output failures now surface a toast. When an AI generation exhausts its retries (network errors, malformed model output, schema validation failures), the platform now shows a sanitized error toast instead of silently resolving to no result. Affected widgets include AI suggestions, AI-assisted drafting, and structured-extraction utilities across modules.