Backslash launches graph to map AI agent endpoint risk

Backslash Security has launched the Agentic Fabric Graph, a tool designed to help security teams assess risks linked to AI agents on employee devices.

The graph provides an interactive view of the software components that make up what Backslash calls the agentic AI fabric on workstations and laptops, including agents, MCP servers, models, skills, hooks and connectors. It is designed to show how those components are linked and where a compromised or overprivileged element could have the widest impact across an organisation.

The launch reflects growing concern among security teams over the spread of AI agents on employee and developer endpoints. While companies may know staff are using AI tools, they often lack a clear picture of how those tools connect to internal systems, local networks, credentials and sensitive data.

Backslash said the graph moves teams beyond basic inventory lists by providing a visual, filterable map of AI-related software on user devices. It is intended to help organisations identify unapproved models, excessive permissions, private account use on company machines and hooks that may enable data exfiltration.

Visibility has become a broader concern in enterprise security. Backslash cited findings from the 2026 SANS AI Survey showing that 63% of respondents identified a lack of visibility into where AI models are used and what they expose.

Endpoint exposure

Backslash positions the graph as a way to answer operational questions that security teams may struggle to resolve quickly. These include which AI agents are deployed across workstations, which users have specific tools installed, which MCP servers are most widely used and whether any of those servers expose the local network.

It also aims to help teams assess the potential effect of risky components by showing the reach of a skill, hook or connector within a company environment. In practice, that means organisations can see which systems, users or endpoints could be affected if a component is misconfigured or compromised.

The graph is part of Backslash's broader platform for securing AI activity on endpoints. According to the company, the platform covers discovery, governance and live protection for AI coding tools, cowork agents and other connected components operating on employee devices.

Backslash said the platform uses intent-based analysis to inspect agent behaviour in real time and block actions such as prompt injection, data exfiltration, privilege escalation and malicious activity by compromised agents. The new graph adds a visual risk layer by helping teams understand how those risks connect across endpoints.

Governance focus

The product comes as enterprises face pressure to bring AI use under clearer internal controls. Security leaders are increasingly being asked not only to detect unauthorised software, but also to understand how AI tools interact with corporate data, code repositories, internal applications and user credentials.

One challenge is that AI agents are not isolated applications. They often rely on external models, local tools, prompts, skills and MCP servers, creating chains of access that can be difficult to document through conventional endpoint management systems. That interconnected structure can leave companies exposed if a single component has broad permissions or hidden network access.

Backslash said the graph is designed to support prioritisation rather than simply report a long list of findings. By highlighting the most widely used components and the violations with the greatest potential impact, it aims to help security teams decide where remediation will matter most.

The company is also offering an agentless, read-only risk assessment based on the graph for enterprises with more than 200 endpoints. Backslash said the assessment maps agents, MCPs, skills and hooks across endpoints, then produces a visual report with findings and a remediation plan.

Shahar Man, Co-Founder and Chief Executive Officer of Backslash Security, said the rise of agentic AI on desktops has created a new class of endpoint risk for enterprises.

"Agentic AI usage is skyrocketing, creating a new interconnected ecosystem of AI agents, MCP servers, models, skills, hooks, prompt rules, and connectors on developer and employee desktops - the agentic fabric," Man said. "This has created new, often hidden paths to credentials, internal systems, code, and business data. The Backslash Agentic Fabric Graph brings full visibility to this new and growing attack surface that is largely unseen by security teams."