Skip to main content

Release Notes — May 2026

Levo Team
Levo Team
Product & Engineering

Release period: 2026-05-01 → 2026-05-31

May deepens Levo's AI security stack: a redesigned Policy Hub with identity-aware policies, a next-generation AI traffic tagger that attributes traffic to individual agents and sessions, and a full Agent Detail view. We also added Streamable HTTP transport for the MCP server, deeper API discovery from source code, and broad reliability and performance gains across the platform.

Highlights​

  • Policy Hub with identity-aware AI policies — The AI Policies page is now a redesigned Policy Hub with inline Monitor/Enforce toggles, server-side pagination, and a new Identity Policy type that matches principals across clouds. Default policies expanded from 16 to 36.
  • AI Agent Detail and per-agent attribution — A new Agent Detail page (Models, MCP, Tools, Findings, Traces) backed by a next-generation AI traffic tagger that attributes traffic to individual agents and reconstructs sessions, including AWS Bedrock AgentCore.
  • MCP server Streamable HTTP transport — The Levo MCP server adds Streamable HTTP transport with header-based auth and new tools, including full vulnerability detail with HTTP traces.
  • Enforced AI guardrails — The AI Gateway wires rate-limit policy actions to token-bucket enforcement, scans LLM response bodies, and enriches guardrail alerts with caller identity and policy lineage.
  • Faster, more reliable platform — Up to 7x faster app and endpoint listing, plus extensive hardening against stalls, deadlocks, and trace drops.
  • Discover more APIs from source — Source-code scanning now runs in parallel across entire GitHub organizations, with offline spec generation and large-repo controls, to surface APIs that may never appear in live traffic.

What's new at a glance. A one-page map of where May's additions land across the Levo API and AI security platform.

May 2026 — What's new

API Discovery & Cataloging​

New Features​

  • Force-refresh OpenAPI spec — Manually force a refresh of an application's OpenAPI spec from the org details page.
  • Discover APIs from source at scale — Source-code scanning now runs in parallel across entire GitHub organizations, with include/exclude directory controls for large repositories and an offline spec-generation mode, surfacing APIs that may never appear in live traffic.
  • Endpoint and application reports now include tags, and applications show all source types when they have both observed traffic and imported endpoints.
  • Source-code scanning skips unsupported languages, classifies empty-spec outcomes, and ignores node_modules and build directories.

Resolved Issues​

  • Endpoint renames fail silently.
  • Environment filtering for application listing does not work correctly.
  • Soft-deleted environments cannot be revived reliably.
  • Storage-version mismatches and dependency errors in the .NET source-analysis path.

API Security Testing​

New Features​

  • AI-authored tests (early access) — New scaffolding and validation APIs for AI-authored tests, with a persisted plan and approval state machine, and test execution gated behind approval status.
  • Exported test plans now produce a runnable zip, and test results are richer with real-time streaming.

Resolved Issues​

  • False positives and weak detection for OS command injection and input validation.
  • LFI tests mishandle empty response bodies.

Web Application Scanning (DAST)​

New Features​

  • Advanced authentication strategies — DAST scans add opt-in multi-step DOM-driven login, session-transplant and storage-state strategies, and static MFA pass-through, plus an iframe-aware login resolver for Keycloak, Auth0, and Okta SSO.
  • Redesigned Create DAST Scan modal — A cleaner scan-creation experience, with the option to run scans on Levo Cloud or on-prem.
  • Time-based SQLi detection with N-sample confirmation, and improved crawler diversity.

Resolved Issues​

  • High false-positive rates across passive and active scanners — reduced by up to ~75% on some paths.
  • DAST scan creation is allowed when the feature flag is disabled.

AI Security​

New Features​

  • Policy Hub with Identity Policies — The AI Policies page is redesigned as a Policy Hub with an Identity Policy editor, inline Monitor/Enforce toggles, and server-side pagination. A new Identity Policy type matches principals across clouds, and default policies expanded from 16 to 36, seeded automatically for new environments.
  • Enforced guardrails and rate limiting — The AI Gateway wires rate-limit policy actions to token-bucket enforcement, scans LLM response bodies for output policies, and supports hot-reload policy enforcement via a new control plane.
  • Richer guardrail alerts — Alerts now carry caller identity (source address, IAM principal, user agent, model, upstream), policy lineage, real scanner names, and gateway-supplied titles, with enforcement mode following the originating policy.
  • Azure OpenAI native passthrough for transparent API-key forwarding, plus guardrail model cache pre-warming and reduced gateway memory usage.

Resolved Issues​

  • Guardrail alerts are not written to a dedicated alerts table.
  • Scanners fire more than once for the same event.

AI Discovery​

New Features​

  • Next-generation AI traffic tagging — A new AI Traffic Tagger attributes traffic to individual agents and sessions, recognizes AWS Bedrock AgentCore runtime callers, and adds A2A discovery and comprehensive AI-provider parsers.
  • Function-calling tools and AgentCore identity — Discover function-calling tools alongside MCP tools, with per-agent identity from AgentCore, and support for server-less MCP tool rows.
  • AI and MCP trace ingestion and querying via a trace-type filter, with a Trace Type switch on the Traces page and PII type names shown inline in the Findings table.

Resolved Issues​

  • MCP entities are modeled incorrectly when captured at the proxy or gateway.
  • Per-agent attribution is incorrect, and agent identifiers are unstable across batches.

Vulnerabilities & Findings​

New Features​

  • Persistent filters — Vulnerability filters are retained across list and detail views.

Resolved Issues​

  • Auto-closing a vulnerability on a flagged endpoint omits your comment.
  • Bulk vulnerability updates are not processed reliably.

Sensitive Data​

New Features​

  • Sensitive data findings for AI workloads — New sensitive-data finding services with entity-type awareness, and PII data pivoted to AI-aware entities so findings cover AI agents and sessions.
  • Entity type is now included across all sensitive-data APIs.

Runtime Protection​

New Features​

  • Protection module health check is now enabled by default.

Sensors & Deployment​

New Features​

  • Java agent published to a public container registry — The Java agent is now available as a published container image.
  • Higher-throughput, self-diagnosing capture — The Java agent adds an async dispatcher with sampling, connection-level sticky sampling, host/port filters, TCP 5-tuple recovery across Tomcat, Jetty, Netty, and Vert.x, per-endpoint rate limiting (dark-launch), and self-diagnosing capture for third-party APIs.
  • The eBPF sensor adds richer socket-info capture and preserves in-progress request data, with noisy health probes suppressed at the capture layer.

Resolved Issues​

  • Silent capture loss and over-capture on Java 8.

Integrations​

New Features​

  • Slack notifications — Levo can now send notifications to Slack.

Resolved Issues​

  • User organization invitations fail.
  • Unaccepted org invites appear in the org picker.

Reporting & Compliance​

New Features​

  • New reports added and page-size handling improved, with a wider Actions column to fit Preview and Download buttons.

Resolved Issues​

  • Report pagination and download issues.

MCP Server​

New Features​

  • Streamable HTTP transport — The Levo MCP server adds Streamable HTTP transport at /mcp with header-based authentication.
  • Full vulnerability detail tool — A new MCP tool returns full vulnerability details including the complete HTTP trace, alongside additional test-authoring tools.

Platform, Administration & Access​

New Features​

  • Polished on-prem sensor management — The Sensors page adds status pills, a Type column, search, compact layout, and contextual empty states, plus a Deployments and Configuration tab.
  • Faster listings — App and endpoint listing is up to 7x faster thanks to new indexes, with broad hardening against stalls, deadlocks, and trace drops for a more reliable experience.

Resolved Issues​

  • Expired sessions are not handled gracefully. A one-shot token refresh now applies, and benign auth-page network errors are quieter.
  • Application name changes do not reflect in filter dropdowns across modules.
  • Stale JavaScript chunks cause load failures.