The Cyber Toolchain — issue 063, September 3, 2026

Strix adds openrouter/z-ai/glm-5.3 as the new top recommended model, alongside GLM-5.3, Fable 5.1, and Gemini 3.7 Flash added to the recommended model list. Users can switch models by setting the STRIX_LLM environment variable, e.g. export STRIX_LLM="openrouter/z-ai/glm-5.3" alongside LLM_API_KEY, then running strix --target ./app-directory.

$ export STRIX_LLM="openrouter/z-ai/glm-5.3"
export LLM_API_KEY="your-api-key"
strix --target ./app-directory

A Bitbucket Access Sync section in team settings imports Bitbucket Cloud user groups and Bitbucket Server groups as ZeroPath teams, with optional hourly auto-sync and a Re-sync now/Sync now button for one-off imports; a permission mapping editor warns when a custom mapping grants fewer permissions to a higher VCSVCSVersion control system: software that tracks changes to files over time, letting teams collaborate and roll back mistakes. Cyber tools use it to manage source code, configs, and signatures. role than a lower one, and unmatched Bitbucket group members appear in the same manual-linking panel used for GitHub, tagged with a platform badge. This extends to a broader access-sync system importing teams from both GitHub and Bitbucket, mapping GitHub's five roles and Bitbucket's three onto thirty-nine permissions across ten groups, and restricting member visibility to only repositories they hold a grant on. Connecting an integration now requires the Create Integrations or Create Installations permission, and adding a repository by URL requires the Create Repositories permission.

A hosted remote MCP endpoint at https://zeropath.com/mcp uses OAuth 2.1OAuth 2.1A consolidated draft of the OAuth authorization framework that merges OAuth 2.0 with its most important security BCP updates, dropping unsafe flows so tools can delegate user access without handling credentials directly. over the Streamable HTTPStreamable HTTPAn MCP transport protocol that lets a client send JSON-RPC requests over HTTP POST while the server streams responses back using Server-Sent Events, replacing the older stdio and SSE transports for remote tool communication. transport, letting clients like Claude, Cursor, and Codex connect via browser authentication with no install or API key; connections are read-only until changes are approved. A local stdio transport mode running the open-source Python server with API-token credentials remains available for scripted or self-managed setups.

$ # In Cursor, Claude, or any Streamable HTTP MCP client, set the MCP server URL to:
https://zeropath.com/mcp
# Then authenticate via the browser prompt that appears — no API key needed.

$ https://zeropath.com/mcp

A per-repository chat-based AppSec Assistant answers security questions and executes automated actions on findings; before complex or irreversible actions (opening pull requests, bulk-updating findings) it proposes a reviewable plan listing Steps, Risks, and scope that must be approved, forceable via a Plan first toggle on any message and savable as a schedule, trigger, or playbook. Plans surface Scheduling and Trigger Suggestions (a cron expression, event trigger, or playbook activation) when a request is recurring or reactive. Organization-specific guidance is authored via the Skills tab (Early Access), with four built-in forkable skills for triage, dependency reachability, report writing, and patch review, and active skills are listed in the plan's skills used field.

Fix PRs for approved patches show progress through Queued, Submitting PR, Retrying, Submission failed, Cancelled, and PR opened states, with specific failure reasons (e.g. missing push credentials) and a Retry button for unrecoverable failures; up to five automatic submission attempts are made before a PR is marked stopped, and Cancel/Retry actions appear directly on the issue.

When enabled, ZeroPath runs a full scan on every meaningful commit to a repository's default branch (including merged pull requests) and incrementally re-examines only changed code on follow-up scans, automatically closing findings resolved by those commits; configurable per organization, tag, or repository in scanner settings, and available on GitHub, GitLab, Bitbucket Cloud, and Bitbucket Server.

Four organization-level report cards are added, visible only to users with organization-level permission: Monthly Security Report (DOCX, with an AI-generated executive summary plus per-repository scan activity, severity breakdown, and top vulnerability families), Organization Summary Report (PDF), SLA Compliance Report (PDF, tracking which findings are within or past configured remediation windows), and SOC 2 (PDF, compliance-focused metrics and scan activity structured for audit evidence). All reports carry a Scope field on the cover page or header showing either All repositories (N) or the specific tags/repositories selected at generation time; by default a report covers the entire organization.

A new Issues By PR Author chart groups open issues from each repository's latest merged pull request scan by author, excluding unmerged PR scans and full-scan-only findings. A new Exposure Window chart shows the average days open issues of each severity were exposed per period, with severity lines breaking rather than dropping to zero when none are open. The date range picker adds a Last 1 year preset and Quarterly granularity alongside daily/weekly/monthly options. The Top Issues widget and scored-findings tables sort unscored findings to the bottom and support up to 500 findings with severity, score, title, repository, vulnerability class, status, and detection-date columns; clicking any row opens a side pane with full issue details, and actions like marking false positive refresh the report in the background without losing filters, scroll position, or selection.

A distinct Issue Tracker Export permission group in team settings covers Create Jira tickets from findings and Create Linear tickets from findings, separate from Issue Management. Jira and Linear ticket links are now included in CSV exports and the export endpoint, and an on-demand Jira ticket sync preview shows what a linked ticket will do before it's applied. Creating Jira and Linear tickets is now a per-repository permission that must be explicitly granted or ticket creation is blocked.

Adds Microsoft Teams integration: paste an incoming webhook URL into ZeroPath to receive scan notifications and vulnerability alerts as Adaptive CardsAdaptive CardsAn open-source card schema from Microsoft that lets apps render structured UI snippets — buttons, text, images — natively inside any host surface, so a tool can push rich, interactive notifications without owning the renderer. in Teams channels, with no Azure AD app registration or tenant administrator required.

The feedback list gains filter chips to narrow by disposition type; the Pending review disposition label is renamed to Needs review, and the Non-exploitable disposition now immediately removes a finding from the open issues queue (same as False Positive or Accepted Risk), whether set by a human or an automated agent.

ZeroPath adds Gerrit change scanning via poll-based triggering (no webhook required): it imports Gerrit projects for full scans, scans each open change as new patch sets arrive, and posts findings as inline review comments on diff lines / as a review message on the patch set.

A new dependency findings tab on the Issues page supports search by CVE, advisory, and package; completed pull request scans link to their findings list with a separate dependency link, and the check run's Details link now points directly to the dependency view when a scan finds only SCASCASoftware composition analysis, identifying third-party dependencies and checking them for known vulnerabilities and license issues. It covers software a team uses rather than writes itself. findings, avoiding a dashboard view that appears empty.

ZeroPath automatically cancels in-progress PR scans for older commits when a newer commit is pushed to the same pull request, across GitHub, GitHub Enterprise Server, GitLab, Bitbucket, Azure DevOps Services, and Gerrit.

Codacy adds Oxlint as a supported engine for JavaScript and TypeScript, covering correctness, style, and best-practice rule categories including ESLint core rules, typescript-eslint, React ecosystem (react, react-hooks, react-refresh, react-perf), @next/eslint-plugin-next, eslint-plugin-unicorn, testing frameworks (jest, vitest), eslint-plugin-vue, import, jsdoc, jsx-a11y, n (Node), and promise. It supports repository-level configuration via .oxlintrc.json, .oxlintrc.jsonc, oxlint.config.json, or oxlint.config.jsonc files, which Codacy detects and applies automatically during analysis.

Adds a read-only permission option when creating an Agent (CI) Web API token, allowing data reads without write access.

Adds a Require exception requests setting to policies, letting teams set developer triage permissions between no triage and full triage and require approval before a blocking finding can be ignored.

Enterprise-managed settings add a model key, configurable per team in team-mappings.json, to set a default Copilot model for each enterprise team; the key is marked overridable so teams can choose their own preferred model. A global enterprise default model applies to all Copilot users, with non-mapped users inheriting that default, and the feature is now generally available across Copilot Business and Copilot Enterprise in the GitHub Copilot app, Copilot CLI, and Visual Studio Code.

⤢

Noir now records source line numbers for route declarations so findings point to the exact line, covering Rails config/routes.rb, OpenAPI/Swagger operations, Symfony #[Route] attributes and routes.yaml, Vert.x, and Spring reactive RouterFunction routes.

OpenAPI documents over 10 MB now get their own 64 MB parsing budget instead of being silently dropped, as they previously were.

JFrog Zero-Touch Remediation automatically evaluates all available fixes for a vulnerability in a customer's environment and selects the best one based on policy, environment, and risk tolerance. It works through Compliant Version Replacement, swapping a vulnerable package version for a patched one mid-pipeline by evaluating multiple upstream sources (Chainguard, Maven Central, and others) against the customer's own policy, with JFrog Artifactory as the system of record.

New APMAPMApplication Performance Monitoring — a category of software that collects metrics, traces, and logs from running applications so developers can detect slowdowns and diagnose their root cause. Repository provides a governed registry for agent package installs, replacing open-internet pulls via apm install. New Plugins Registry delivers a scanned, auditable source for first- and third-party coding agent plugins, replacing default public marketplaces. New Agent Package Resolution provides a trusted, policy-enforced path for agents to resolve dependencies through the JFrog Platform.

New JFrog Agent Plugin integrates coding agents — Claude Code, Cursor, VS Code Copilot, Kiro, Devin, OpenCode, and Codex — natively with the JFrog Platform.

Adds an AI-Powered Policy as Code Playground that converts plain-language intent into enforceable, versioned, Rego-based policy automatically, requiring no manual policy authoring.

Adds token_type and remediation fields to ExtraData in the GitHub V2 detector, differentiating between PATPATA personal access token, a credential used by scripts and integrations instead of a user's password. Its permissions and expiry depend on the service issuing it. (ghp_, github_pat_), OAuth (gho_), and GitHub App (ghu_, ghs_, ghr_) token types so incident responders are directed to the correct revocation path.

Adds a V2 detector for Weights & Biases, enabled via the WeightsAndBiasesV2DetectorEnabled feature flag, expanding credential coverage for the ML platform.

Adds the SOURCE_TYPE_AWS_SECRETS_MANAGER source type, enabling TruffleHog to scan AWS Secrets Manager for leaked credentials.

Adds --bypass-end-user-auth flag to fleetctl package to generate fleetd installers that skip end-user authentication prompts on Linux and Windows hosts (requires fleetd v1.60.0 or higher). Adds mdm.allow_orbit_end_user_auth_bypass server setting, enabled by default, to control whether hosts that skip end-user authentication can enroll into a fleet that requires it; set to false to strictly enforce authentication for all Orbit enrollments.

mdm:
  allow_orbit_end_user_auth_bypass: false

$ fleetctl package --type=msi --fleet-url=https://fleet.example.com --enroll-secret=<secret> --bypass-end-user-auth

Adds PATCH /api/v1/fleet/fleets/{id} support for {"features": {"enable_software_inventory": <bool>}} to enable or disable software inventory collection per fleet; omitting the key leaves the stored value unchanged.

$ curl -X PATCH https://fleet.example.com/api/v1/fleet/fleets/7 \
  -H 'Authorization: Bearer <token>' \
  -H 'Content-Type: application/json' \
  -d '{"features": {"enable_software_inventory": false}}'

Adds support for $FLEET_HOST_VITAL_<id> custom host vital variables in Android configuration profiles, managed app configurations, and host name templates, with per-host expansion and automatic re-delivery when a vital value changes.

Adds support for setting minimum_version to latest with deadline_days in macOS/iOS/iPadOS OS update policy to automatically track the newest Apple-published version per device (Fleet Premium).

Adds s3_software_installers_signed_url configuration option to serve software installers, in-house apps, and bootstrap packages via GCS presigned URLs so clients download directly from object storage instead of routing through the Fleet server.

Adds token_invalid field to ABM token API responses and dep_device_error (human-readable message) to GET /hosts/:id/dep_assignment to surface why an Apple Business Manager device lookup or ABM token is failing. Also adds support for releasing devices from Apple Business Manager inside Fleet.

Adds 29 new iOS/iPadOS device vitals (battery level, accessibility settings, cellular technology, cloud backup status, MDM options, device attestation, and more) collected via the existing DeviceInformation MDM refetch and exposed in the host API response and a new View all vitals modal; BYODBYODBring Your Own Device: a policy allowing employees to use personal phones, laptops, or tablets for work, creating a network of unmanaged endpoints that cyber tools must monitor or enforce policy on. enrollments are excluded.

Adds host_id and host_serial fields to the mdm_enrolled activity for Apple (macOS, iOS, iPadOS) enrollments, and surfaces the activity on the host's activity timeline.

Adds support for Fleet built-in variables (e.g. $FLEET_VAR_HOST_END_USER_IDP_USERNAME) in software install, post-install, and uninstall scripts, resolved per host at execution time (Fleet Premium).

Adds a default fleet assignment for new Windows MDM enrollments (Fleet Premium), applied before the Autopilot Enrollment Status Page runs so software, scripts, and configuration profiles take effect during out-of-box setup.

Adds user_mfa_requested activity recorded when valid credentials are submitted for an MFA-enabled account and a verification email is sent. Adds created_setup_experience_script and deleted_setup_experience_script activities to the audit log when a setup experience script is added, replaced, or removed via the API or GitOpsGitOpsA practice where Git repositories serve as the single source of truth for infrastructure and application state, with automated systems continuously reconciling live environments to match what is declared in version control..

Adds per-fleet activity automations (Fleet Premium): a webhook configured from the Hosts page that posts to a destination URL whenever an activity tied to one of the fleet's hosts is created.

Adds Adobe Creative Cloud plugin detection (CEP and UXP extensions) to software inventory on macOS and Windows, listed as software type Plugin (Adobe) with version and host count.

New Charlotte AI AgentWorks lets security teams describe a desired security outcome in natural language and build, test, and refine custom agents grounded in Falcon telemetry, with no coding required. Security workflows such as triage, enrichment, hunting, and response can then be run conversationally within Anthropic Claude via AgentWorks, with results returned inline and full auditability.

Administrators can enable or disable scheduled report failure notifications via Administration > Notifications. By default, a daily failure digest email is sent to the master administrator and the report schedule creator whenever scheduled reports fail, including the report name, report ID, schedule ID, failure time, and failure reason. Notifications also surface clear failure codes and descriptions, widget-level error details for PDF report failures, and configuration-specific guidance such as reminders to verify required frameworks and asset groups.

A new ack = true option on the accept_splunk operator enables the HECHECAn HTTP API endpoint provided by Splunk that lets applications stream JSON-formatted events directly into a Splunk index without a forwarder agent, reducing pipeline complexity for log-shipping tools. acknowledgement protocol: clients receive an ackId per request, queryable via /services/collector/ack, with acknowledgements held for 10 minutes by default and delivery guarantees preserved across node crashes/restarts.

$ accept_splunk hec_token = secret("splunk-hec-token"), ack = true

The parallel operator now accepts fuse (values: none, parallel, or all) and limit_partitions options, matching the // parallelism directive and capping keyed operator parallelism for operators like summarize and deduplicate.

$ parallel 8, fuse = "none", limit_partitions = 2 {
  where y > 100
  summarize g, total = sum(x)
}

The new read_avro operator decodes streams of concatenated raw Apache Avro binary datums (not Avro Object Container Files), and the new parse_avro function decodes a single raw Avro datum from each blob or string value; both require the writer schema via the schema argument.

$ read_avro < data.avro | ...

OpenAI's Advanced Account Security (AAS) program now requires hardware-backed passkeys for all Trusted Access for Cyber (TAC) members, disabling weaker phishable methods such as SMSSMSShort Message Service, the protocol built into cellular networks for sending text messages between phones; cyber tools use it for alerting, two-factor authentication delivery, or phishing simulation over mobile channels. OTPs and push notifications upon security key enrollment. Enrolling a YubiKey under the AAS program at chatgpt.com/advanced-account-security covers both ChatGPT and Codex accounts with hardware-backed passkeypasskeyPhishing-resistant sign-in credentials built on FIDO2/WebAuthn: the device holds a private key and proves possession instead of the user typing a password. They sync across a user's devices. protection.

The My Organization API and Embeddable UI Components now expose self-service configuration of third-party application access for an Organization and for individual connections, plus the Cross App Access (XAA) Resource Application role for a connection, so Organization Admins can manage access without routing changes through the Auth0 Dashboard. Connections gain a third_party_client_access object (default { "default_value": "block", "allowed_values": ["block"] }) and settings for User Access Authorization and Client Credential Access Authorization, each configurable as Unauthorized, Authorized, or All. XAA configuration carries a domain-verification guardrail — the connection's identity provideridentity providerA service that authenticates users and supplies identity information to other applications. It is the sign-in authority behind many single sign-on systems. must use a domain verified for the associated Organization — and each delegated setting is independently controllable so platform teams choose which settings are exposed to Organization Admins.

⤢ ⤢ ⤢ ⤢

The jcam check_config command now validates jcamAgencyDefinedDataItems mappings against the attributes JCAM actually carries, reporting entries configured for attributes JCAM does not have and attributes JCAM has that are not mapped.

Automated Control Testing now tests linked evidence records automatically as they are submitted, delivering a pass/fail control-level evaluation with detailed rationale and immediate gap flagging instead of waiting for the next manual audit cycle.

Highlights wrapped in == can now be given a color by prefixing the text with a color emoji (e.g.) or by using the new color picker in the formatting submenu. Typing == now triggers an inline suggestion popup for highlight colors, and Live Preview shows a color swatch when the cursor overlaps a highlight, which can be clicked to change its color.

iOS 26 Native Quick Capture lets you capture notes from the Lock Screen, Control Center, or Shortcuts without waiting for your vault to load. It supports configurable Create Locations — capture to a new note, today's daily note, a bookmarked note, or any chosen note — and an active Quick Capture note surfaces as a Live Activity on the Lock Screen and Dynamic Island while editing.

Bases now supports a kanban layout for displaying files in ordered, drag-and-droppable columns, defined via the Group by setting.

Adds an optional SettingDefinitionItem#id field for developers to set a unique ID on individual settings, allowing stable references when settings share a name or a setting page is renamed.