Fileless execution on Linux has moved from niche tradecraft into real-world intrusion chains. By executing payloads from memory or anonymous file descriptors, attackers can reduce on-disk artifacts and weaken controls that rely heavily on file inspection.In our own analysis of VoidLink, we observed how fileless execution can be paired with a rootkit. The loader scans for processes running from memfd and passes their PIDs to the rootkit, allowing an already running fileless implant to be hidden…
Elastic Defend 9.5.0 detects dynamic link library (DLL) search-order hijacking [1] in a single field. Writing that rule before 9.5.0 took about 88 lines, covering approximately 2,600 named libraries, 10 excluded Windows system paths, signature checks, and a drop-to-load time window. It now takes one: dll.Ext.defense_evasions: "DLL Hijack: Masquerading".DLL search-order hijacking, which Defend labels Masquerading, runs attacker code inside a legitimate process by abusing the order that Windows…
AI verdict correctness in our security operations center (SOC) is 92%, up from 60%, but we didn't switch models to get there. What we changed is the context the agents get before they decide anything, including the detection rule's investigation guide and user risk data from Workday, along with the closure reasons from 30 days of past cases on that same rule. This post covers how the agentic SOC pipeline is built in Elastic Workflows and Elastic Agent Builder, down to the prompts and the…
AI verdict correctness in our security operations center (SOC) is 92%, up from 60%, but we didn't switch models to get there. What we changed is the context the agents get before they decide anything, including the detection rule's investigation guide and user risk data from Workday, along with the closure reasons from 30 days of past cases on that same rule. This post covers how the agentic SOC pipeline is built in Elastic Workflows and Elastic Agent Builder, down to the prompts and the…
We gave hundreds of developers an AI agent that can run shell commands, edit files, and call Model Context Protocol (MCP) servers on their laptops, then realized we had no record of what it actually did. So we built one. One 280-line dependency-free bash script, fired by Cursor's hooks, records every tool call as JSONL, and the Elastic Agent already on each endpoint ships it to Elasticsearch. Since the May rollout we have logged over 13 million tool-call events from more than 1,100 machines. A…
npm's min-release-age setting tells npm to ignore any package version published less than a set number of days ago, keeping freshly compromised releases out of npm install during the window when they do the most damage. Getting the setting onto developer workstations is straightforward. Knowing when someone quietly deletes it is a different problem entirely, and log-tailing inputs are no help because they only fire when lines are appended to a file. We built a ~40-line Common Expression…
An agentic SOC is only as good as the model driving it. The moment you let an LLM triage an alert, hunt across your telemetry, or author a detection rule, the question stops being "is this a smart model?" and becomes something much more specific: will it pick the right skill, call the right tool in the right order, and reach the right disposition without inventing a result it never actually checked? That is not a question a general-purpose leaderboard can answer. A model can top every public…
An agentic SOC is only as good as the model driving it. The moment you let an LLM triage an alert, hunt across your telemetry, or author a detection rule, the question stops being "is this a smart model?" and becomes something much more specific: will it pick the right skill, call the right tool in the right order, and reach the right disposition without inventing a result it never actually checked? That is not a question a general-purpose leaderboard can answer. A model can top every public…
Large language models (LLMs) made it trivially cheap to generate vulnerability reports. In the first half of 2026 alone, our HackerOne bug bounty program received over 1,390 reports, more than the full-year totals for 2024 and 2025 combined. Every one of them still requires human attention, so we decided to put agents against agents. If AI can generate reports at near-zero cost, AI should triage them at near-zero cost, too. The system we built agrees with human security engineers 85% of the…
Large language models (LLMs) made it trivially cheap to generate vulnerability reports. In the first half of 2026 alone, our HackerOne bug bounty program received over 1,390 reports, more than the full-year totals for 2024 and 2025 combined. Every one of them still requires human attention, so we decided to put agents against agents. If AI can generate reports at near-zero cost, AI should triage them at near-zero cost, too. The system we built agrees with human security engineers 85% of the…
Hugging Face reconstructed more than 17,000 attacker events from a July 2026 intrusion driven by an autonomous artificial intelligence (AI) agent. The path was familiar: untrusted dataset content abused a processing worker (file disclosure, then code execution), credential harvest, then multi-cluster lateral movement. Production Elastic Defend behavior rules and Elastic Security detection (SIEM) rules already watch those types of behaviors. This post maps each stage to detections you can enable…
This is Part 3 of the Inside Elastic InfoSec's Agentic SOC series. Part 1: How we triage every alert before an analyst opens it · Part 2: Choosing the right agent architecture for a 5× cost reduction We run 14 AI agents in the Elastic InfoSec security operations pipeline. They were producing correct verdicts and taking up to 19 large language model (LLM) calls to do work that needed 8, at thousands of input tokens per call. At hundreds of runs per day, that compounds fast. We built a five-step…
This is Part 3 of the Inside Elastic InfoSec's Agentic SOC series. Part 1: How we triage every alert before an analyst opens it · Part 2: Choosing the right agent architecture for a 5× cost reduction We run 14 AI agents in the Elastic InfoSec security operations pipeline. They were producing correct verdicts and taking up to 19 large language model (LLM) calls to do work that needed 8, at thousands of input tokens per call. At hundreds of runs per day, that compounds fast. We built a five-step…
This is Part 2 of the Inside Elastic InfoSec's Agentic SOC series. Part 1: How we triage every alert before an analyst opens it. Part 3: how we cut AI agent LLM calls by 60%. Investigating a Windows endpoint alert in Elastic InfoSec's production agentic security operations center (SOC) costs $0.69. That's what we pay running an orchestration workflow of specialized Elastic AI agents on the Elastic Inference Service (EIS). Route the same alert to a single agent working through 14 skills, and the…
This is Part 2 of the Inside Elastic InfoSec's Agentic SOC series. Part 1: How we triage every alert before an analyst opens it. Part 3: how we cut AI agent LLM calls by 60%. Investigating a Windows endpoint alert in Elastic InfoSec's production agentic security operations center (SOC) costs $0.69. That's what we pay running an orchestration workflow of specialized Elastic AI agents on the Elastic Inference Service (EIS). Route the same alert to a single agent working through 14 skills, and the…
We ran a noisy wget detection rule on Elastic's own cloud fleet for seven days. Three destinations survived deterministic filtering, Elasticsearch Query Language (ES|QL) COMPLETION triaged all three, and none of them created an alert that an analyst had to open. Each rule parses the destination from curl and wget executions, filters known-good hosts, redacts secrets, and then hands whatever’s left to a large language model (LLM) for a triage verdict. File transfer detections stay on in cloud…
This is Part 1 of the Inside Elastic InfoSec's Agentic SOC series. Part 2: choosing the right agent architecture for a 5× cost reduction. Part 3: how we cut AI agent LLM calls by 60% Elastic's InfoSec team built an agentic SOC that triages every alert before an analyst opens it. A 30-minute manual investigation now finishes in under 3 minutes: deterministic ES|QL queries close obvious false positives at zero token cost, specialized AI agents investigate the rest across endpoint, cloud, and SaaS…
This is Part 1 of the Inside Elastic InfoSec's Agentic SOC series. Part 2: choosing the right agent architecture for a 5× cost reduction. Part 3: how we cut AI agent LLM calls by 60% Elastic's InfoSec team built an agentic SOC that triages every alert before an analyst opens it. A 30-minute manual investigation now finishes in under 3 minutes: deterministic ES|QL queries close obvious false positives at zero token cost, specialized AI agents investigate the rest across endpoint, cloud, and SaaS…
Elastic's InfoSec Product Security Team built a generative AI agent using Elastic Agent Builder that drafts complete CVE security advisories (CWE classification, CAPEC methodology, CVSS scoring, and mitigation guidance) directly from raw vulnerability reports. The agent uses RAG against the MITRE CWE and CAPEC catalogues indexed in Elasticsearch, which grounds its output in authoritative data and prevents hallucinated classification IDs. ESA-2026-01 is already in production as an example of…
AAD Graph Activity Logs are now ingestible into Elastic and usable for threat detection within the SIEM/XDR solution. That sentence shouldn't be exciting, but it is. For most of the past decade, this slice of telemetry simply didn't exist as a customer-accessible log stream. Microsoft Graph Activity Logs (the modern graph.microsoft.com surface) went GA in April 2024. The legacy graph.windows.net surface, the one adversary tooling actually hits, stayed dark until early 2026. This post walks the…