Release History
What's new in Behavry. This page highlights the product and platform changes most relevant to operators and security teams; internal infrastructure and tooling changes are omitted.
June 2026
Ask BehavryAI — grounded analyst copilot. The assistant widget became a read-only analyst copilot that can explain alerts, escalations, agents, risk scores, and policy decisions using your tenant's live data. It runs a tool-use loop over a read-only catalog, forces tenant scoping on every query, and only ever returns secret-free summaries (finding types, never raw payloads or DLP match values). Advisory only — no write actions. Activates when the AI endpoint/key/model are configured.
Sessions becomes the live-execution hub. Activity (live decision feed) and Behavior (monitoring) now fold in as tabs alongside the Sessions timeline, with a restructured primary nav (Overview · Governance · Runtime · Discovery · Operations · Administration) and collapsible drawers for lower-tier tools.
Agent Provenance Records (APR). A new evidence layer assembled from the audit stream:
- Evidence packages — exportable, signed bundles for a session or investigation.
- Delegation graph + behavioral trajectory — visualize how authority and behavior flowed across an agent's session.
- Narrative generation + external KMS signing — human-readable incident narratives with cryptographic signing.
- Chain status UI + notifications/escalations surfaced as first-class records.
Alerts & Escalations promoted to first-class Runtime surfaces, with their own pages rather than living inside other views.
AI Exposure & Discovery — human (browser) governance folded into the AI Exposure surface alongside agent and citizen-app discovery.
Unmanaged MCP server discovery — when an agent calls an unregistered MCP server, Behavry audits, escalates, and blocks (shadow-server detection) instead of returning a bare error.
Data-subject rights (DSAR). Right-of-access export and right-to-erasure (anonymize-in-place, preserving the audit hash chain) for GDPR / CCPA workflows. See GDPR / Data Privacy.
Operations — trusted-IP allowlist (BEHAVRY_RATE_LIMIT_ALLOWLIST) so a burst of failed logins from an operator IP never locks out a plane.
May 2026
Compliance as a live control plane. Framework mappings with live metrics and auditor-ready CSV/JSON/PDF exports across SOC 2, ISO 27001, NIST AI RMF, PCI DSS, HIPAA, GDPR/Data Privacy, EU AI Act, AIUC-1, and OWASP ASI. See Compliance.
Audit proof path hardening. Ed25519 detached signatures on audit events, a public verification endpoint, and a signed export manifest with an offline verifier — so a third party can confirm an exported decision trace hasn't been tampered with.
Controlled Evidence Access. Forensic workflows with an explicit access ceremony for disclosing protected payload content, each disclosure itself audited.
Per-dimension behavioral baselines. The Behavioral Monitor learns separate baselines per metric (frequency, error rate, resource use, volume), improving anomaly precision.
MCP Supply-Chain Integrity. CVE matching and TLS certificate pinning for target MCP servers, surfaced in the dashboard.
Adversarial Red Team Swarm (ARS). ML-driven attack synthesis and multi-stage attack chains feeding the policy-automation loop.
Audit Chain + Storage Health, including an audit-chain re-baseline flow to record a reason and reset a degraded chain state after an operational break.
Looking for an older entry? The full historical changelog is maintained in the repository's CHANGELOG.md. This page is curated for product-facing changes.