Backslash

No entries match these filters. Clear filters

1. announced Oct 26, 2026 · issue 123

   Backslash: Plugin severity now comes from the highest-risk nested child, so a low-risk wrapper can't hide a critical embedded script or tool manifest.

2. AI Plugin Inventory: Uncover Hidden Risks | Backslash - Backslash (2026-10-26) Oct 26, 2026 · issue 122

   Backslash adds a native AI Plugin Inventory with child-component risk scoring, signature verification, and inline host-level guardrails for agentic endpoints.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   • Adds a Plugin Inventory (Plugins tab) that discovers every AI plugin installed on developer endpoints — surfacing publisher, source repository, unpacked contents, risk severity, associated agents, and last sync timestamp, fully filterable and sortable — alongside child-component risk scoring that assigns a plugin's overall severity by the highest risk found among its nested scripts, MCP tools, hooks, and rules (preventing a low-risk wrapper from concealing a critical child component), and MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server and agent-skill mapping that ties every MCP connection and skill to its parent plugin and host agent, surfacing over-permissive tool definitions and untrusted content sources.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Adds a Hooks Inventory that automatically discovers hooks across Claude Code, Codex, Cursor, and Copilot, tracking whether a hook was added by Backslash, a developer, or a third-party plugin, combined with deep code parsing and digital-signature validation for hook files to verify component provenanceprovenanceEvidence recording where a software artifact came from and how it was built, including its source and build process. It helps a consumer check that the artifact matches a trusted build. and detect unauthorized script tampering before execution, while vetting scripts against backdoor and credential-theft threat models.
   • Adds local model discovery on macOS and Windows model directories for runtimes such as Ollama, LM Studio, and llama.cpp, governing locally-run models under the same allow/deny controls as remote foundation models.
   • Adds AI Rules vetting that scans standing configuration files (.cursorrules, AGENTS.md, CLAUDE.md) for instruction poisoning and exposed credentials, with guardrails that automatically quarantine unsafe files at session start.
   • Enforces inline host-level guardrails that block unapproved command executions, credential access, and data egress before actions execute — rather than alerting after the fact.

   +1 more show less

   • Delivers automated audit records and host-level run tracing aligned to regulatory frameworks including EU AI ActEU AI ActEU regulation that classifies AI systems by risk level and imposes compliance obligations—documentation, transparency, human oversight—on developers and deployers, shaping how AI-based cyber tools are built and sold in Europe., NIS2NIS2EU directive that expands mandatory cybersecurity requirements to more sectors and member states than its predecessor, giving cyber tools a compliance framework to map controls and reporting obligations against., DORA, and SOCSOCA security operations center, the team or function that monitors threats, investigates alerts and coordinates incident response. Its work can be internal, outsourced or shared., without requiring manual developer reporting.

   6 details in Everything

3. announced Oct 19, 2026 · issue 116

   Backslash adds Local Model Discovery: local AI models on endpoints enter the AI inventory with allow/deny controls, permission-gap reporting, and deduplication.

   • New supported platformDiscovers local AI models on macOS and Windows endpoints and adds them to the AI inventory alongside remote models, with the same allow and deny controls.
   • Reports a permission gap when a model directory cannot be read, distinguishing an access limitation from an empty directory.
   • Deduplicates multiple paths pointing to the same model into a single inventory record and filters out files that merely resemble models.

   3 details in Everything

   • New integration Local AI tool discovery integrations
4. Local Model Discovery: Govern Shadow AI | Backslash - Backslash (2026-10-19) Oct 19, 2026 · issue 114

   Backslash adds Local Model Discovery to surface shadow AI from Ollama, LM Studio, llama.cpp, and six other local runtimes on macOS and Windows.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ HOW TO FIND IT
   Review all local models discovered across enterprise endpoints and identify any policy violations before applying allow/deny controls.
   📍In Backslash, go to AI Inventory › filter by local designation › review each model's risk details and policy violations › add to allow or deny list.
   • Adds Local Model Discovery to detect AI models installed on macOS and Windows endpoints via Ollama, LM Studio, Jan, GPT4All, the Hugging Face cache, MLXMLXApple's machine-learning framework for Apple silicon, which uses the unified memory shared by CPU and GPU to run and fine-tune models on a Mac., LocalAI, and llama.cpp — with no developer setup required.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Surfaces local models in the existing AI inventory alongside remote models, enabling the same allow and deny list controls to be applied to endpoint-resident models.
   • Reports a permission gap when a model directory cannot be read, letting security teams distinguish an access limitation from an inspected directory that contained no models.
   • Deduplicates multiple paths pointing to the same model into a single inventory record and filters out files that merely resemble models.
   • Ties each discovered local model to its specific endpoint even when no AI agent is attached to it.
   • Adds Local Model Discovery to detect AI models installed on macOS and Windows endpoints via Ollama, LM Studio, Jan, GPT4All, the Hugging Face cache, MLXMLXApple's machine-learning framework for Apple silicon, which uses the unified memory shared by CPU and GPU to run and fine-tune models on a Mac., LocalAI, and llama.cpp — with no developer setup required.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Reports a permission gap when a model directory cannot be read, letting security teams distinguish an access limitation from an inspected directory that contained no models.
   • Deduplicates multiple paths pointing to the same model into a single inventory record and filters out files that merely resemble models.
   • Ties each discovered local model to its specific endpoint even when no AI agent is attached to it.

   1 more in Everything

5. AI Fabric Graph: Connected Agent Security | Backslash - Backslash (2026-10-13) Oct 13, 2026 · issue 109

   Backslash launches AI Fabric Graph, connecting enterprise endpoints, AI agents, MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers, skills, and plugins in one risk-investigation view.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ HOW TO FIND IT
   Investigate a policy violation starting from a specific endpoint: open the AI Fabric Graph, select the endpoint, filter for violations, and follow the chain to the offending agent or MCP server to pinpoint the root cause.
   📍In the Backslash console, open AI Fabric Graph › select an endpoint › enable the violations filter › click the flagged agent or MCP server node › open the details panel to view the policy violation and affected components.
   Determine how broadly a risky MCP server or skill is shared across the enterprise before deciding on a remediation policy.
   📍In the Backslash console, open AI Fabric Graph › select a component (e.g. an MCP server or skill) › inspect the connected agents and endpoints listed in the graph to see every installation that inherits the risk.
   • Adds the AI Fabric Graph, a connected visualization that chains enterprise endpoints, AI agents, MCP servers, skills, rules, and plugins into one view with risk severity and Shadow AI indicators.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Enables filtering the graph for policy violations and drilling into agent-level violation details to identify the root cause and affected components.
   • Supports tracing a flagged asset back to every agent and endpoint connected to it, revealing shared exposure across the agentic fabric.
   • Allows teams to start from a component and enumerate which agents and endpoints share it, enabling scoped remediation across multiple installations.

   4 details in Everything

6. announced Oct 13, 2026 · issue 109

   Backslash: Security teams can start an investigation from an endpoint or component, filter violations, and trace flagged assets to all connected agents and endpoints.

   └──▷ HOW TO FIND IT
   Investigate a suspicious MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server component to see every agent and endpoint that connects to it and assess how broadly a finding extends across the environment.
   📍In the Backslash console, open the AI Fabric Graph, select a component (e.g. an MCP server), filter for policy violations, then follow the connections to the agents and endpoints that share the component.
7. Claude Code mods: What they are, and what you should be aware of - Backslash (2026-10-12) Oct 12, 2026 · issue 107

   Backslash flags Claude Code mod risks: use claude plugin validate, allowManagedModsOnly, and ~/.claude/skills/ audits to govern agent plugins.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ TRY IT
   Inspect a mod before installing it to see which events it hooks and which Claude Code functions it calls.
   copy

   $ claude plugin validate ./my-mod-folder

   Lock a managed team environment so only company-approved mods can run — preventing unknown third-party mods from loading.
   yaml copy

   allowManagedModsOnly: true

   List all installed plugins (including auto-loading ones in ~/.claude/skills/) to audit what is active in the current environment.
   copy

   $ claude plugin list

   • Run claude plugin validate on a mod's folder to list the events it hooks and the functions it calls before installing.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Enable allowManagedModsOnly in company policy so only company-deployed mods can run in Claude Code sessions on managed machines.
   • Audit ~/.claude/skills/ periodically — plugin folders placed there load automatically in every Claude Code session with no install step.
   • Use /plugin inside a Claude Code session, or claude plugin list from the shell, to enumerate all active and installed mods including those in ~/.claude/skills/.
   • Backslash provides a single-pane view across developer endpoints showing every installed agent plugin, MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server, skill, and hook alongside what each can reach and which carry risk.

   5 details in Everything

8. Block MCP action seen Jul 23, 2026 · issue 023

   Backslash adds a Block MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. action to control MCP activity.

   • New Block MCP action enables practitioners to block MCP (Model Context ProtocolModel Context ProtocolModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface.) actions within Backslash.

   1 detail in Everything

9. Claude Code support seen Jul 23, 2026 · issue 023

   Backslash adds support for Claude Code.

10. PR Event Notifications seen Jul 23, 2026 · issue 023

    Backslash adds PR event notifications to alert teams of security findings at pull request time.

11. Remove PR comment after successful scans seen Jul 23, 2026 · issue 023

    Backslash now removes PR comments automatically after scans come back clean.

    • Automatically removes previously posted PR comments when a subsequent scan returns no findings, keeping pull requests free of stale security noise.

    1 detail in Everything

12. Selected repos shown in top seen Jul 23, 2026 · issue 023

    Backslash now surfaces selected repositories at the top of the repo list for faster navigation.

    • Selected repositories are now pinned to the top of the repository list for quick access.

    1 detail in Everything

13. New MCP Spec Opens Three New Attack Surfaces. Security, Get Ready. - Backslash (2026-05-28) May 28, 2026 · release notes

    Backslash introduces endpoint monitoring for new MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. stateless protocol attack surfacesattack surfacesThe exposed systems, interfaces and access points an attacker could try to use. It includes internet-facing services as well as identities, software and other reachable resources..

    • Adds visibility into browser_id and other handle strings passed in conversation context to detect potential hijacking.

    1 detail in Everything

14. AI Agent Skills: The New Standard for Modular AI Workflows - Backslash (2026-04-15) Apr 15, 2026 · release notes

    Backslash introduces Agent Skills for modular AI workflows and a free Skills Security Scanner

    • Enables teams to package organizational knowledge and specialized workflows as portable SKILL.md files for use across Claude Code, GitHub Copilot, VS Code, and other agents.
    • Supports installation of skills via registry command npx skills install <skill-name> or by manually placing skill folders in project (~/.github/skills/), personal (~/.codex/skills), or system-level directories.

    2 details in Everything