Skip to content

What's new

Changes since the previous documentation update (27 August 2026), newest areas first.

October 2026

Enhanced Dashboard

A redesigned dashboard is now the default for every account, organized around the questions a risk team gets asked. See The Enhanced Dashboard.

  • Overview with posture tiles, framework readiness as "N of M controls evidenced", trends, value and impact, a five-step readiness setup wizard, and presentation mode for board meetings
  • Needs Attention: alerts, risks, reviews and deadlines in one queue, worst first
  • Model Register with a dossier per system: verdicts you can mark reviewed, signals, model card, control coverage, trust score history and audit trail
  • Assurance: explainability (now including DiCE counterfactuals), fairness and drift, each with a plain-language verdict
  • Observability: performance, LLM & RAG audit trails, agents and evaluations
  • Compliance & Evidence: framework detail for all five frameworks, control applicability decisions, evidence filing and approval, Collect system evidence, evidence packs, a compliance calendar and an audit trail with an integrity check
  • Reviews & Decisions: requests, boards, the decision archive and RACI in one place
  • Lenses ("View as"), a ⌘K command palette, keyboard shortcuts and saved views
  • A Support page with an in-app request form

The Classic layout remains available for now from the avatar menu.

Compliance

  • Control catalogs for ISO/IEC 42001 (63 controls), the EU AI Act (28, following the Digital AI Omnibus amendments), GDPR (25), CCPA (22) and NIST AI RMF (19 categories). See Compliance frameworks.
  • ISO/IEC 42001 registers: record AIMS policies with an approval step, and measurable AI objectives, from the dashboard.
  • NIST AI RMF detail by core function, with open risks per function.
  • Access reviews in Settings → Security posture: keep or revoke each account, with notes, recorded as attestations.
  • Model card refresh from the dossier, pulling the latest metrics, drift and bias results.
  • New risk register categories: hallucination and privacy. Changing severity thresholds now re-rates every risk.

Reports

The Report Studio replaces the old report engine: 12 report types in PDF, Word or HTML, templates with writing style and glossary, multiple brand profiles, Preparer / Reviewer / Approver sign-off, Verify a file, auditor share links with presets and watermarks, and schedules that send a secure link. See Reports.

Governance

  • Votes and finalizations need a current authenticator code, and are refused if the request changed since the voter opened it. Organization API keys can't vote.
  • Review requests take attachments and typed discussion comments.

Evaluations and prompts

  • Golden datasets, deterministic scoring (exact match, required phrases, rule-based checks), and a regression check between prompt versions. No LLM-as-a-judge.
  • Governed prompt versions with a design → verify → deploy → retired lifecycle, rollback, and links from each LLM call to the version that produced it.

See Evaluations & Prompt Versions.

Alerts

Alert rules are now evaluated automatically when metrics are logged, drift reports are created and bias audits complete. See Needs Attention.

Policy-as-Code & GitOps

Sync a whitebox.yaml policy from your repository and evaluate a model against it to warn or block a deployment from CI. See Policy-as-Code & GitOps.

Pricing

Plans are now sized in Governed AI Systems and Governance Events instead of API calls. Free includes 1 system and 2,500 events a month; Business Cloud up to 10 systems and 100,000 events, with overage at $5 per 10,000. Dashboard use, reads, configuration and reports are free. See Plans & Limits and Governance Events.

Python SDK 1.3.0

  • governance_run() to group a run's records into shared Governance Events
  • client.prompts, client.compliance and client.reports
  • Model version comparison (compare_versions, version_changelog, list_git_versions)
  • DiCE counterfactual parameters on explanations
  • client.governance.get_review_request and submit_decision
  • Clearer rate-limit handling: long Retry-After waits are raised rather than slept through

See Python SDK and LLM monitoring.

MCP server 0.3.0

  • 197 tools across 18 domains, including prompts, compliance, model diff and alert triage
  • MCP Apps interactive cards for explanations, alert triage, governance dossiers, ISO/IEC 42001 status and model version diffs
  • API-key-only local authentication, and a hosted endpoint at https://api.whiteboxxai.com/mcp/
  • Governance voting from the dossier card, with a person's own session and authenticator code

See MCP Server.

For AI agents

Seven public agent skills, a compliance-assistant prompt, and a llms.txt index of these docs.