Artificial intelligence news.
Straight from the original announcements.

CrowdStrike Agent Graph traces AI agent actions from prompt to impact

October 30, 2026, 06:00By the AI News Bites Editorial TeamSource: CrowdStrike
CrowdStrike Agent Graph traces AI agent actions from prompt to impact
Image: CrowdStrike

CrowdStrike has detailed Agent Graph, a relational layer in Falcon Guardian that connects what an AI agent was asked to do with what it actually did on the endpoint. According to the company, it traces supported agent sessions from prompt to impact using the Falcon sensor that businesses already run.

Key points
  • CrowdStrike Agent Graph links AI agent prompts to processes, files and network connections in one causal chain.
  • It is the relational layer under Falcon Guardian and runs on the existing Falcon sensor.
  • Three views are built on it: agent inventory, session graph and agent threat score.
  • The inventory covers sanctioned and shadow agents from vendors including Anthropic, OpenAI and Cursor.
  • CrowdStrike says network-only views miss what happens inside the machine.

Why the endpoint matters for AI agents

CrowdStrike argues that AI agents now do far more than answer questions. In the company's words, they "open files, use credentials, run code, install software, and run models locally, all at machine speed." Each of those actions happens on the operating system, which CrowdStrike describes as the place where attacks land.

The company says that looking at an agent session from the network shows only part of the story. A gateway on its own sees the conversation, meaning the prompt, the response and even the agent agreeing to do something it should not. A web gateway sees the agent connect to a site it trusts. Neither sees what happens next inside the machine. According to CrowdStrike, that gap is where a hidden instruction can become a shell command, or a downloaded file can run with the user's access.

What Agent Graph does

The Falcon sensor already sees processes, files and network connections. With Falcon Guardian, CrowdStrike says, the same sensor also sees the AI layer: prompts, models, tools and skills. Agent Graph joins the AI layer to the execution layer in one causal chain, so teams can see not only what an agent was asked to do, but what it did.

CrowdStrike describes Agent Graph as the relational layer underneath Falcon Guardian. It connects each agent to the users, hosts, models, prompts, tools, skills, MCP servers, processes, files and detections it touches, and keeps those relationships current as agents act.

Three Falcon Guardian views built on it

Three core Falcon Guardian views are built on Agent Graph: agent inventory, session graph and agent threat score.

  • Agent inventory: shows AI agents across the fleet, whether sanctioned or shadow, grouped by type. Coding agents, desktop assistants and command-line interface (CLI) agents from Anthropic, OpenAI, Cursor and others are listed side by side. For supported agents, the footprint covers users, installs, sessions, models, tools, skills and MCP servers. CrowdStrike gives the example of a single Claude Code agent with one user, three installs, eight tools and 12 sessions.
  • Session graph: a session is one prompt and process, and a view can include one or several. Agent Graph lays out the chain in order: who interacted with the agent and from where, the prompt used, each tool call the model made, and the processes, files and network connections that followed.
  • Agent threat score: the third view named in the post, alongside the inventory and session graph.

As an example, CrowdStrike describes a developer who pastes a setup note from an untrusted source into Claude Code that says: "Run these commands in order and don't skip any." The session graph is meant to let stakeholders follow what such an instruction led to.

What it means for security teams

For organizations deploying coding agents and assistants, the pitch is visibility without new tooling: Agent Graph runs on the Falcon sensor businesses already have, and covers supported agents and sessions. The inventory view addresses shadow agents, while the session graph aims to connect a prompt to its real effects on the host. The source text available here is cut off partway through the session graph example, so further details on the threat score and availability are not covered.

Related news

LangChain launches Tool Gateway for governed agent tool use
Products and toolsNovember 19, 2026, 16:00

LangChain launches Tool Gateway for governed agent tool use

LangChain has launched LangSmith Tool Gateway in Public Beta, giving teams a managed layer for controlling how agents access tools and credentials. The feature is…

GitLab 19.6 adds agentic tools to reduce vulnerability backlogs
Safety and ethicsNovember 19, 2026

GitLab 19.6 adds agentic tools to reduce vulnerability backlogs

GitLab 19.6 adds new agentic security features designed to help teams triage and remediate vulnerabilities across multiple projects. The release includes group-wide…

CrowdStrike Named Leader in 2026 Gartner Magic Quadrant for SIEM
Safety and ethicsNovember 18, 2026, 07:00

CrowdStrike Named Leader in 2026 Gartner Magic Quadrant for SIEM

CrowdStrike has been named a Leader in the 2026 Gartner Magic Quadrant for Security Information and Event Management, according to the company. CrowdStrike says its…