<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>FairMind changelog</title><link>https://fairmind.xyz/changelog/</link><description>What shipped in FairMind, release by release.</description><item><title>FairMind v0.5.0: Agents that explain themselves</title><link>https://fairmind.xyz/changelog/#v0.5.0-alpha</link><guid isPermaLink="false">v0.5.0-alpha</guid><pubDate>Sun, 22 Nov 2026 11:19:33 GMT</pubDate><description>Your AI systems and agents can now send FairMind their live runs, with the reason they gave at each step. FairMind checks those reasons by replaying decisions, and counts the result toward the release gate. This release also adds ISO 42001 internal audits and frameworks you follow.
Live runs: A running AI system or agent sends its runs (model calls, tool calls, retrievals, memory, decisions) with the reason it stated at each step, through the SDK's `record_trace`. Tool-call monitoring and groundedness evaluations read them, nothing to upload, and the release gate counts them for monitoring.
Explained decisions: Explanation quality replays recent decisions on the model connection, never the agent's tools. It shows what drove each decision, whether the stated reason held, and the smallest change that reverses it. It counts for human oversight, Autopilot can rerun it, and every result is kept under Records.
Shadow AI agent: FairMind as osquery's server with edge AI model detection, a laptop account check, MCP and model events, rules, a Rust menu bar and tray notifier, a browser account check for Chrome and Edge, and an early-access kit.
Internal audit (ISO 42001 9.2): Audit plan, areas, multiple auditors and a sealed trace. Its work shows under Needs you.
Frameworks you follow, and an auditor portal built for the auditor . AI-drafted, framework-specific scoping that a person confirms.
Evaluation runs count in the release gate and ISO clause status by one rule everywhere. They are listed under Evidence, where reviewers can map passed ones to a control.
Evaluation correctness: Judge verdicts, restricted probes, fairness point estimates and strict upload parsing.
Autopilot runs catalogue behaviour tests on a chat model. Agent: Needs attention flags only real failures.
Pilot gaps closed: Clause evidence, SoA export, two-person risk acceptance, autopilot lock, clause map, assurance pack and execution.
Discovery Found list, and Home lists systems with no evidence once.
Change an AI system's owner from the system file, recorded on the change chain.
Task reminders for work due tomorrow or overdue, with unsubscribe.
Consent through Olai for every signed-in user and every purpose FairMind's collection point asks.
The sidebar footer shows who is signed in; sign-ins are recorded. MCP client files with missing tools or system prompt are treated as not read.
Docs aligned with the code, maturity levels labelled, dead surface removed.
Prod deploys `main` automatically from a timer on the instance.
Dependency updates.</description></item><item><title>FairMind v0.4.2</title><link>https://fairmind.xyz/changelog/#v0.4.2-alpha</link><guid isPermaLink="false">v0.4.2-alpha</guid><pubDate>Mon, 16 Nov 2026 21:30:43 GMT</pubDate><description>`fairmind eval run &lt;id&gt; --system &lt;id&gt; --local`: Run behaviour tests (prompt injection, jailbreak, leakage, over-refusal, output bias) on models that live on a private network. Set the system's connection to "A runner inside your network"; the CLI then calls the model from inside your network and sends only the answers back, so nothing needs inbound access.
`fairmind audit &lt;framework&gt; --system &lt;id&gt;`: One command to prepare a framework audit, such as ISO 42001. Add `--local` to run its behaviour tests through the same runner.
Shadow AI: Seat lists and laptop inventories merged per laptop; Defender, CrowdStrike and Chrome sources; a laptop script for signed-in accounts, laptop MCP servers, Fleet and proxy rules; cited explanations from the org's own model.
Tasks: Assign work to people, or to an AI agent that does it and reports back.
Discovery: Removing a source works again and asks first; each source shows how much of the account its last read covered and what it found. vLLM and described ChatGPT and Claude workspaces as sources. Personal GitHub accounts are refused as discovery sources.
Organisations are matched by email domain, not typed name. Org roles: owners are protected and an org always keeps one owner.
Frameworks section redesigned around a requirement map and trace.
System page: "Run an evaluation" opens the evaluations that run, and the model connection is labelled as such.</description></item><item><title>FairMind v0.4.1</title><link>https://fairmind.xyz/changelog/#v0.4.1-alpha</link><guid isPermaLink="false">v0.4.1-alpha</guid><pubDate>Mon, 16 Nov 2026 08:51:10 GMT</pubDate><description>Re-release of v0.4.0-alpha so Zenodo can archive it and mint a DOI. Zenodo failed on v0.4.0-alpha because CITATION.cff listed two licences; it now names the repository licence, AGPL-3.0-only. The SDK, CLI and MCP server remain Apache-2.0. No code changes since v0.4.0-alpha; see its notes for everything in this release, including the corrected FairMind-E confidence policy.</description></item><item><title>FairMind v0.4.0</title><link>https://fairmind.xyz/changelog/#v0.4.0-alpha</link><guid isPermaLink="false">v0.4.0-alpha</guid><pubDate>Mon, 16 Nov 2026 08:26:56 GMT</pubDate><description>Discovery and Shadow AI, change and retirement records, a security pass, and the corrected FairMind-E confidence policy.
The platform is now AGPL-3.0; the SDK, CLI and MCP server are Apache-2.0.
FLOSS/fund `funding.json` manifest.
Confidence is the provenance class midpoint; a caller-supplied score is recorded for audit but never enters the gate.
A power tracker that falls back to a CPU-load model is classed `manual`, not `tool_estimated`.
Unattended power-reference session with five mitigation repeats.
GREENS 2027 camera-ready draft and a constant-free method-disagreement statistic.
Discovery connects vendor admin APIs, clouds, sign-in, gateways, logs and MCP.
Shadow AI, with merged sightings, owner check and decision.
Material changes and retirement are recorded on each AI system.
One governance grid across Home, Frameworks, Evidence, System and Plan.
Side nav redesign and Ask FairMind; Home, the nav badge and Needs you agree.
Release autopilot: Data sources and an agent that runs measurable route steps.
Governance plan with agentic discovery and a board.
Maintenance page while the app restarts.
Fake, placeholder and pass-on-nothing results removed from the app.
Refresh-token rotation, an ESLint gate and a report-only CSP with nonces.
Security headers on every page and no-store on API answers.
Access ends when a member is removed or signs out.
The invitation journey works end to end, with emails, encoded tokens and same-site redirects.
The backend defaults to production; dev login only against a local database.
Migrations run before the backend starts; no `create_all` outside development.
Blocking route handlers and org scope queries run off the event loop.
Org audit log rows are written; governance audit reports are no longer created at request time.
`setup_db` runs on Neon's direct endpoint.
Backend image pinned to Python 3.11.
Sandbox catches OOM kills Docker's flag misses.
Docs rebuilt around the app, page by page, with recordings.
All 109 catalogue evaluations run; engine status derived from the code.
FSE 2027 industry paper data and scripts.
OpenSSF Best Practices badge.</description></item><item><title>FairMind v0.3.1</title><link>https://fairmind.xyz/changelog/#v0.3.1-alpha</link><guid isPermaLink="false">v0.3.1-alpha</guid><pubDate>Sun, 08 Nov 2026 23:00:45 GMT</pubDate><description>The first release published to PyPI and the MCP Registry.
Packages publish to PyPI through trusted publishers, with no stored token, and the MCP server is listed in the MCP Registry on each release.
Package versions now come from the release tag.
Releases carry SLSA build provenance alongside the Sigstore signatures.
The fairness metric computations are fuzzed with Atheris in CI.
Catalogue evaluations can be listed, described and run from the SDK and the MCP server.
Worker job lifecycle: Start, signed completion, failure, and cancellation that wins races.
S3 artifact store for evaluation inputs.
Eval-runner image, sandbox launcher and seccomp profile.
P0 frontend roadmap items closed.
The dashboard framework card shows only core frameworks.
Roadmap rewritten with a feature status table.
The migration ledger is written on a clean install.
PostgreSQL 18 catalog re-measured.
JOSS paper with the required sections, AI usage disclosure and a draft PDF workflow.
Named FairMind-E arXiv preprint source.
README lists the Awesome AI Regulation, Principles &amp; Guidelines listing.</description></item><item><title>FairMind v0.3.0</title><link>https://fairmind.xyz/changelog/#v0.3.0-alpha</link><guid isPermaLink="false">v0.3.0-alpha</guid><pubDate>Sat, 07 Nov 2026 19:44:48 GMT</pubDate><description>FairMind v0.3.0-alpha. **Status: alpha.** See [docs/STATUS.md](https://github.com/adhit-r/fairmind/blob/main/docs/STATUS.md) for what works today. This is the first release with Sigstore-signed packages. Every release now attaches the `fairmind` SDK, `fairmind-cli` and `fairmind-mcp` packages, each with its `.sigstore.json` bundle, and publishes the backend image to `ghcr.io/adhit-r/fairmind-backend`.
Behaviour tests: 20 evaluations that probe a live AI system and judge its answers.
WEAT and SEAT ported to the EV-FAI-05 evaluation engine.
Fairness checks say when they pass only because the gap is not statistically significant.
Policy rules gate release approval, are read live, and count as framework control evidence; policy owners are alerted when a published rule starts failing.
Draft AI policies with the organisation's own model.
Manual evidence capped at 0.30 in the release gate; shortest path to Go priced and ties broken by evidence strength.
First-verdict flow that explains what you are shipping and why.
Risk taxonomy data model (IBM/MIT) and the OECD AI Principles framework catalogue.
Security hardening: Pickle RCE, SQL injection, CI template injection, OAuth secret handling and XSS fixes.
Backend trust controls, granular permissions and a leased worker queue on PostgreSQL.
Frontend trust slice: Session shell, scope isolation, evidence gates and quarantine.
Gate evidence PATCH on stored URL metadata; duplicate creates on retry stopped.
12 of 13 open CodeQL alerts cleared.
SQLAlchemy 2.1 and psycopg 3; FastAPI 0.162 with lazy router inclusion.
Dependabot for Python, npm, Docker and GitHub Actions; actions pinned by SHA and images and pip installs pinned by hash.
OpenSSF Scorecard workflow and badge.
Boundary, archive-import and frontend audit checks in CI.</description></item><item><title>FairMind v0.2.0</title><link>https://fairmind.xyz/changelog/#v0.2.0-alpha</link><guid isPermaLink="false">v0.2.0-alpha</guid><pubDate>Thu, 05 Nov 2026 14:59:40 GMT</pubDate><description>First archived release of FairMind, the open-source AI assurance workbench for models and agents. **Status: internal alpha.** The evidence contract, trust and review foundations are built and tested. Sandboxed evaluation workers and validated LLM, agent and multimodal evaluation packs are not shipped yet. See [docs/STATUS.md](https://github.com/adhit-r/fairmind/blob/main/docs/STATUS.md) for what can and cannot be claimed today. FairMind does not provide compliance certification. This release adds machine-readable project metadata (CITATION.cff, codemeta.json) and is archived on Zenodo for citation.</description></item><item><title>FairMind v2.0.0-rbac: 🚀 RBAC System Complete &amp; Production Ready</title><link>https://fairmind.xyz/changelog/#v2.0.0-rbac</link><guid isPermaLink="false">v2.0.0-rbac</guid><pubDate>Mon, 23 Mar 2026 18:07:00 GMT</pubDate><description>5 RBAC Tables: Organizations, org_members, org_invitations, org_roles, org_audit_logs.
9 API Endpoints: Member management, invitations, roles, audit logs, compliance reports.
Data Isolation: 5 enforcement layers (middleware, query decorators, row-level security).
Email Workflows: Registration requests, member invitations via Resend.
Permission Decorators: @require_org_owner, @require_org_admin, @require_org_member.
Fine-Grained Permissions: Custom permissions per role with @require_permission decorator.
Stacking Decorators: Compose multiple permission checks for complex authorization.
Global &amp; Org-Scoped Roles: Support for both organization-level and global roles.
Authentik Provider: RS256 token signing, PKCE flow, state parameter protection.
Frontend Auth: Callback handler at /auth/callback with JWT token storage.
Token Validation: RS256 signature verification, claim extraction with org_id.
ComplianceReportGenerator: JSON/CSV/PDF report generation.
Audit Logging: Track all org operations: user, action, IP, timestamp, details.
Compliance Dashboard: Metrics cards, events timeline, action distribution charts.
Audit Report Endpoint: GET /api/v1/organizations/{org_id}/compliance/audit-report.
Registration Form: Neobrutalist design with role selection.
Invitation Acceptance: /invitations/[token] with 5-state machine.
Organization Switcher: Multi-tenant org selection.
Compliance Dashboard: 5 React components for metrics, charts, tables.
64 Backend Tests: Multi-org isolation, security, performance benchmarks.
32 Frontend E2E Tests: Full user journey with Playwright.
CI/CD Pipeline: GitHub Actions automated testing.
Performance Verified: All critical operations &lt;500ms.
7 Comprehensive Guides (20,000+ words):.
Architecture overview with ASCII diagrams.
API reference with curl examples.
Permission system matrix.
Deployment guide with checklist.
Operations playbook.
50+ FAQ entries.
Complete navigation index.
Bold 4px black borders throughout.
6-8px hard geometric shadows.
Orange (#FF6B35) accent buttons.
High contrast, no gradients.
Sharp corners for confidence.
Audit-ready compliance styling.
Backend: FastAPI with async/await support.
Database: Neon PostgreSQL with 19 optimized indexes.
Email: Resend API with verified domain.
Auth: Authentik OAuth2/OIDC provider.
Frontend: Next.js with React components.
DATABASE_URL (Neon).
AUTHENTIK_URL.
RESEND_API_KEY.
JWT_SECRET.
Registration approval workflow.
Invite acceptance rates.
API usage by org/role.
Email delivery success.
Src/api/routers/compliance.py (NEW).
Src/services/compliance_report_service.py (NEW).
Tests/test_*.py (5 new test files).
Src/app/(auth)/invitations/[token]/page.tsx (NEW).
Src/app/(dashboard)/compliance/dashboard/page.tsx (NEW).
Src/components/compliance/* (NEW, 5 components).
Tests/e2e/rbac.spec.ts (NEW).
Docs/* (7 new markdown guides).
[x] Backend API functional.
[x] Database migrations prepared.
[x] Email service configured.
[x] Frontend OAuth2 integration.
[x] Permission decorators.
[x] Audit logging.
[x] Comprehensive tests.
[x] Complete documentation.
Multi-tenant data isolation.
OAuth2/OIDC authentication.
Comprehensive compliance support.
Full test coverage (96 tests).
Production-ready documentation.</description></item></channel></rss>