Changelog

Changelog

All notable changes to AegisGate Security Platform are documented here.

v3.3.0-beta.2 - 2026-06-08 - EU AI Act Beta (Integrity Hotfix) 🩹

Released 2026-06-08. This is a code-content fix, not a security fix. v3.3.0-beta.1 was tagged on the same day, but a release-integrity review discovered that 5 commits implementing the EU AI Act work (Phases 1.1, 1.2, 1.3) and the gitignore enforcement work had never been merged to main. v3.3.0-beta.2 fixes this integrity gap. No CVE, no vulnerability, no leaked data. The missing work was always present on a feature branch; it was never in any public release.

What changed vs. v3.3.0-beta.1:

• ✅ EU AI Act sub-package (pkg/compliance/eu-ai-act/, 82 controls) — NOW ACTUALLY INCLUDED (was missing from beta.1)
• ✅ EU AI Act customer 1-pager (docs/compliance/eu-ai-act.md, 104 lines) — NOW INCLUDED
• ✅ EU AI Act full control mapping (docs/compliance/eu-ai-act-mapping.md, 438 lines) — NOW INCLUDED
• ✅ Marketing content files (content/pricing.md, content/tech.md, content/changelog.md) — NOW INCLUDED
• ✅ Website update (docs/website/index.html, EU AI Act section) — NOW INCLUDED
• ✅ Pre-commit guard rail (.githooks/pre-commit blocks plans/ + legal-docs/ staging) — NOW ACTIVE
• ✅ .gitignore policy header (12-line “no force-add” directive) — NOW IN EFFECT
• ✅ gofmt — 4 files cleaned up (CI caught 4 files needing formatting in the merge)
• ✅ Action versions — all 8 actions in .github/workflows/security.yml bumped to latest (Node 20 → Node 24, fixes GitHub deprecation warnings)
• ✅ gitleaks CI re-enabled — free OSS license configured as GITLEAKS_LICENSE GitHub secret (was disabled in beta.1 due to paid license requirement)
• ✅ 15 P3/P4 golangci-lint issues fixed (in commit 8b69aa2, pre-beta.2)

Tiers, modules, and pricing — unchanged from beta.1.

For full release notes see CHANGELOG.md in the GitHub repo and the v3.3.0-beta.2 release page.

v3.3.0-beta.1 - 2026-06-08 - EU AI Act Beta (Superseded)

Note: This release was superseded by v3.3.0-beta.2 on the same day (2026-06-08) due to a release-integrity gap: the EU AI Act sub-package and related work were advertised in the CHANGELOG but had not been merged to main at the time of the v3.3.0-beta.1 tag. The v3.3.0-beta.1 tag is preserved at SHA 64d0ab5 for historical record; do not use it. Use v3.3.0-beta.2 instead.

Released 2026-06-08. This is a BETA release, not a commercial launch. Professional+ tier and module buy buttons are intentionally hidden pending counsel review and the v3.4.0 paid pentest. Starter and Developer tiers remain live and sellable.

Highlights:

• EU AI Act Compliance Module (NEW) — 82 controls across 8 categories of the EU AI Act, gated by the existing license + module framework. Available to Professional+ tier via the (hidden) module buy button.
• 7-tool security self-attestation — gosec, govulncheck, golangci-lint, gitleaks (837→0 findings via allowlist), trivy, syft, nmap. Self-attested as production-ready for beta; full pentest planned for v3.4.0+.
• v3.3.1 hardening pass — Dockerfile base images pinned to SHA256 digests, K8s seccompProfile.RuntimeDefault added, all 16 P1+P2 golangci-lint findings fixed.
• Minimum-viable legal kit — 6 v2.0 customer-facing docs + 1 Beta User Agreement, all published at /legal/. Self-drafted, not legal advice; counsel review deferred to v3.4.0+.
• Buy-Button Visibility — Professional tier (2 buttons) and all 6 module buy buttons are hidden in the v3.3.0-beta.2 web UI with a “coming-soon” placeholder. The 4 Starter + Developer buttons remain live.

Available tiers in this beta:

• Community (free) — unchanged
• Starter ($29/mo or $290/yr) — Buy Button live
• Developer ($79/mo or $790/yr) — Buy Button live
• Professional ($499/mo or $4990/yr) — hidden pending v3.3.0-GA
• Enterprise (custom) — contact sales

Module availability:

• All 6 modules (HIPAA, PCI-DSS, SOC 2, ISO 42001, FedRAMP, FIPS 140-2/140-3) are hidden in this beta. EU AI Act is the only new module addition; the other 6 were already available in v3.2.0 but are now gated behind the same hidden placeholder.

Verified:

• go build ./... exit 0
• go test ./... 49 packages PASS, 0 FAIL
• gitleaks 0 findings (allowlist in .gitleaks.toml)
• golangci-lint 15 P3/P4 issues (deferred)
• Security CI: 8 jobs all green
• Release workflow: binary + container + SBOM attestation all successful

For full release notes see CHANGELOG.md in the GitHub repo and the v3.3.0-beta.1 release page (superseded — see v3.3.0-beta.2 above).

v3.2.0 - In Progress (Phases 1–5 complete; release prep next)

Released 2026-06-05 (Phase 1, module extraction) — final release after Phase 6.

Trust Framework — 5th Pillar (Phase 4) — Completed 2026-06-05

The newest architectural pillar: continuous, per-agent cryptographic trust scoring with Ed25519-signed attestations. The Trust Framework ties the four protocol pillars (HTTP API, MCP, A2A, RESPONSE) into a single, auditable trust story.

• Per-session trust score — GET /api/v1/trust/score?session=ID returns a 0–100 score with a component breakdown (threat blocks, tool risk average, A2A capabilities used, response findings).
• Signed attestations — every event emits an Ed25519-signed attestation. Verify offline with the public key.
• Attestations feed — GET /api/v1/trust/attestations?since=TIMESTAMP&limit=100 returns the rolling event log.
• Professional+ gate — the TRUST_PILLAR_ENABLED feature flag activates the API for Professional and Enterprise tiers. Lower tiers receive HTTP 402.
• Powers the Compliance Scan Engine — the trust layer is the substrate that the /api/v1/compliance/scan and /api/v1/compliance/report endpoints query.

Compliance Modules (Phase 1, Module Extraction) — Completed 2026-06-05

Six billable compliance modules are now available as add-ons to any paid tier. Prices are locked from the pricing-table decision (2026-06-04) and will not change for existing customers (Q2: lock-in at purchase price forever).

Modules are purchased via Stripe checkout and activated instantly on the customer’s license via the existing webhook (Q1: instant via Stripe webhook). All 6 module products are live in the Stripe dashboard and rendered on the Pricing page.

Pro Tier Price Change (Phase 2) — Completed 2026-06-05

Pro tier is now $499/mo (was $249/mo) and $4,990/yr (was $2,490/yr). The change is in the Stripe dashboard, billing-config.json, the pricing page, and the tier-comparison docs. Existing Pro customers would be grandfathered at $249/mo per the locked decision Q2 (lock-in at purchase price forever); however, no Pro customers exist at the time of this release, so the grandfathering code path is unexercised.

The Pro tier also picked up the Trust Framework and the Compliance Scan API as Professional+ features in v3.2.0 — see the Tier Comparison page for the updated matrix.

Compliance Scan Engine (Phase 3) — Completed 2026-06-05

Two new public endpoints, available on Developer+ tiers (with a valid license key):

• GET /api/v1/compliance/scan — returns a snapshot of every enforced framework: total controls, enforced controls, compliance percentage, and any missing modules that would close gaps. Response shape:

  { "frameworks": [
      { "name": "hipaa", "controls_total": 54, "controls_enforced": 48, "compliance_pct": 88.9, "missing_modules": [] },
      { "name": "pci",   "controls_total": 64, "controls_enforced": 64, "compliance_pct": 100.0, "missing_modules": [] }
  ] }
  

• GET /api/v1/compliance/report?framework=hipaa — returns the full control list for a single framework, with enforced, description, and the source module. Suitable for auditor export.

The scan engine is backed by the trust layer’s attestation feed (Phase 4), so control coverage reflects the actual enforced behavior, not a static catalog.

Website Refresh for the 5-Pillar Story (Phase 5) — Completed 2026-06-05

The marketing site has been updated to reflect the v3.2.0 5-pillar architecture:

• Homepage — the hero card grid now lists the four protocol pillars plus a highlighted Trust Framework card (new in v3.2.0).
• Tech page — the “Five Pillars” section is reordered (HTTP API, MCP, A2A, RESPONSE, Trust) and the Trust Framework gets its own detail table (8 components, 2 API endpoints, attestation format, tier gating).
• Pricing page — Pro tier is now $499/mo and the 6 compliance modules are listed with tier requirements and Stripe buy buttons.
• Tier Comparison — Pro price updated, new “Trust Framework” and “Compliance Scan API” rows added.

Upcoming in v3.2.0

• Release engineering — VERSION bump, GPG-signed annotated tag v3.2.0, GitHub Release with notes, SBOM re-generation, cosign signing, container image push to GHCR. (Phase 6.)

v3.1.1 - Tier Rate Limit Drift Fix

Released 2026-06-05

This release fixes a critical drift between the website-promised tier limits and the code-enforced tier limits. Customers who paid for Starter, Developer, or Professional were silently receiving 40–70% underdelivery on rate limits and concurrent users/agents.

Tier Rate Limit Corrections

• Starter tier — modeled as a first-class tier for the first time (was faked via a starter_mode feature flag with 50% underdelivery)
• Developer tier — proxy 600 → 1000 RPM, MCP 300 → 500 RPM, users 10 → 25, agents 5 → 25
• Professional tier — proxy 3000 → 10000 RPM, MCP 1500 → 5000 RPM, users 50 → 100, agents 25 → 100
• Enterprise tier — unchanged (-1/-1, unlimited)

Webhook Hardening

• Tier validation in the Stripe checkout webhook: unknown values are now rejected with a structured invalid_tier error instead of silently generating a license for an arbitrary string
• Tier alias normalization (e.g., "pro" → "professional", "free" → "community") for consistent downstream behavior

Tooling

• Go 1.26.3 → 1.26.4 (security fix for crypto/x509 and net/textproto stdlib vulnerabilities)
• Binary aegisgate-platform is no longer tracked in git; use go build to compile from source (see README “Build from Source”)

v2.x Status

v2.x is end-of-life as of 2026-12-31. No security updates will be issued after that date. v3.x is the only actively supported line. See the v2.x EOL notice for migration guidance.

v3.1.0 - Current

Released 2026-05-27

New Security Packages

• atlas — MITRE ATLAS adversarial AI threat detection (66 techniques, 100% coverage)
• computeruse — Claude Computer Use security guard (URL denylist, screenshots, keystrokes)
• correlation — Cross-protocol threat correlation engine (94.4% coverage)
• trust/identity — Agent cryptographic identity (ECDSA P-256, 90.9% coverage)

Threat Modeling

• Comprehensive STRIDE analysis for all packages
• CVSS scoring for 25+ threats (7 Critical, 11 High, 7 Medium)
• MITRE ATLAS full coverage (ATLAS-MCP, ATLAS-A2A, ATLAS-LLM)
• Full documentation at /plans/THREAT-MODEL-STRIDE.md

Security Improvements

• Fail-Secure Bridge Design — Insecure fallback replaced with fail-closed behavior
• Prompt Injection Detection — 8 new patterns added (45+ total)
• OWASP LLM Top 10 — Full implementation complete
• MITRE ATLAS — 66 adversarial AI techniques (up from 52)

CI/CD Improvements

• DCO bot detection: EMAIL field check for [bot] patterns
• Build error fix: webhook server return statements
• GOFLAGS=-mod=mod: security workflow dependency listing

Key Metrics

• Test Coverage: 97.8% (up from 97.7%)
• Total Tests: 5,484 (up from 2,548)
• Security Coverage: 92.4%
• Compliance Coverage: 98.7%

v3.0.0

Released 2026-05-20

• ACP Protocol security layer with HMAC verification
• Agent Response Security (RESPONSE package) — PII/secret scanning
• Five-pillar architecture: HTTP API, MCP, A2A, ACP, RESPONSE (the Trust Framework was added as the cross-cutting 5th pillar in v3.2.0)
• 97.7% test coverage, 2,548 tests
• OWASP LLM Top 10 patterns (49 patterns)

v2.0.1

Released 2026-05-06

• FAIL-CLOSED AUDIT: 9 critical + 5 high vulnerabilities fixed
• A2A capability persistence — capabilities survive pod restarts
• Comprehensive health checks — /health verifies all subsystems
• SLA/SLO definitions — new /api/v1/sla endpoint
• gosec alerts resolved — directory/file permissions, error handling

v2.0.0

Released 2026-05-05

A2A Agent Security (Major Release)

• mTLS authentication — X.509 certificate validation for agent identity
• HMAC-SHA256 integrity — Full request body signature verification
• Capability enforcement — Least-privilege per agent from YAML config
• Token-bucket rate limiting — Per-agent request quotas (default 100 req/min)
• License-aware A2A enforcement — ECDSA P-256 cryptographic validation

Observability

• A2A Prometheus metrics (a2a_license_failures_total, a2a_capability_denials_total, etc.)
• A2A startup status logging in server logs

For older releases, see GitHub Releases.