Loading vulnerability…
CVE-2026-58138 is a critical unauthenticated remote code execution vulnerability in Orkes Conductor versions 3.21.21 through versions before 3.30.2. Workflow API endpoints accept attacker-supplied inline workflow definitions before authentication. Malicious JavaScript or Python expressions in INLINE, LAMBDA, DO_WHILE, or SWITCH tasks can reach GraalVM evaluators configured with unrestricted host access, including HostAccess.ALL or allowAllAccess(true). This permits escape from the intended scripting environment through Java reflection or direct subprocess invocation and execution of arbitrary operating-system commands.
Mallory correlates every CVE against your assets, your vendors, and active adversary campaigns. Know which vulnerabilities matter for you, not just which ones are loud.
What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.
What an attacker gets, and what they’ve been doing with it.
If you can’t patch tonight, do this now.
Patch, then assume compromise.
5 valid exploits after Mallory filtered fakes, detection scripts, and README-only repos (2 hidden).
This three-file repository contains an operational unauthenticated remote-code-execution PoC and a Nuclei validation template for CVE-2026-58138 affecting Conductor/Orkes Conductor versions 3.21.21 through 3.30.1. `CVE-2026-58138.py` is a standalone Python command-line exploit using only standard-library modules. It creates a timestamped workflow, POSTs it to the workflow metadata API, starts it via the workflow API, then polls task status up to 15 times and prints the captured command result. Its INLINE JavaScript evaluator payload abuses GraalVM host access and Java reflection to reach `java.lang.Runtime.exec`, launching `sh -c <command>` on the server. `CVE-2026-58138.yaml` is a Nuclei template that performs the same three-request workflow and confirms RCE by matching a randomized marker emitted by `echo {{marker}}-$((6*7))`. `README.md` documents affected versions, impact, mitigations, and detection advice, although its claim that the repository contains only non-destructive detection logic conflicts with both included executable exploit artifacts. No fixed victim hostname or IP is embedded; targeting is operator-controlled through the Python target argument or Nuclei `Hostname` variable.
Repository contains a working exploit for CVE-2026-58138 affecting Orkes Conductor 3.21.21 through 3.30.1. There are 4 files total: a standalone Python PoC, a Nuclei YAML template, a README, and a license. The Python script is the main exploit implementation: it accepts a target base URL and optional command, builds a malicious workflow definition with an INLINE task, and injects a JavaScript payload that abuses GraalVM host access to reflect into Java classes and call java.lang.Runtime.exec(). Exploitation proceeds in three stages: POST /api/metadata/workflow to register the malicious workflow without authentication, POST /api/workflow/{name} to start it, and repeated GET /api/workflow/{id}?includeTasks=true to retrieve the INLINE task result and display command output. The payload is operational rather than just demonstrative because it supports arbitrary attacker-supplied commands and returns stdout. The YAML file is a Nuclei template that automates the same three-step flow for verification, using a marker-based echo command to confirm code execution. Overall, the repository’s purpose is unauthenticated remote command execution against exposed Conductor workflow APIs, not merely detection.
This repository is a reproduction kit for CVE-2026-58138 affecting Conductor OSS. It is not just a scanner rule set: it includes a working Python exploit PoC, OpenTaint rules/models, and a Java test project used to validate the detection logic. The main exploit artifact is poc/poc_inline_lambda_rce.py, a standalone Python script that targets unauthenticated workflow-definition submission in vulnerable Conductor deployments and achieves server-side command execution through GraalVM script evaluators. Repository structure: (1) poc/ contains the live exploit; (2) .opentaint/rules/ contains custom taint rules identifying unsafe org.graalvm.polyglot.Context.eval usage when full host access is enabled; (3) .opentaint/model/config/ contains passthrough models for GraalVM fluent builders; (4) .opentaint/test-projects/ contains a small Gradle Java project with positive and negative samples plus rule-test.yaml; (5) README.md documents vulnerability context, reproduction steps, and affected/fixed versions. Exploit capability: the PoC submits a malicious workflow definition to Conductor’s /api/workflow endpoint. The injected expression is evaluated by vulnerable INLINE, LAMBDA, DO_WHILE, or SWITCH task evaluators. For JavaScript, the payload abuses GraalVM host access by reflecting from a bound Java object ($.getClass().getClass()) to resolve java.lang.Runtime and invoke Runtime.exec with a shell command. For Python, the README/script indicate use of a subprocess-based payload in the GraalPy evaluator. The exploit then polls workflow execution and confirms success either by reading returned stdout from task.outputData.result (INLINE/LAMBDA) or by checking a marker file written under /tmp in the container (DO_WHILE/SWITCH). Targeting: README states the scan and exploit target Conductor v3.23.0 and that the issue is fixed in 3.30.2. The vulnerability is described as unauthenticated RCE caused by code injection into GraalVM polyglot evaluators with HostAccess.ALL or equivalent full-access configuration. Notable endpoints and indicators: the exploit uses a base URL such as http://localhost:8000 and interacts with /api/workflow and workflow-status retrieval paths. The Java test project exposes numerous illustrative Spring @GetMapping routes showing vulnerable and non-vulnerable patterns, but these are for rule validation rather than the real target. Overall, this is a credible, operational exploit repository with both offensive PoC code and supporting static-analysis artifacts.
Repository is a small standalone PoC for CVE-2026-58138 targeting Orkes/OSS Conductor. The main exploit logic is in exploit.py, a Python 3 script using only the standard library. It accepts a target base URL and command, generates a malicious JavaScript expression for an INLINE workflow task, submits the workflow to the Conductor API, starts execution, and then fetches workflow/task output to recover stdout from the executed command. The exploit is clearly operational rather than just demonstrative because it automates the full attack chain end-to-end. Core capability: unauthenticated remote code execution against vulnerable Conductor instances by abusing GraalVM host access in script evaluators. The payload uses reflection from the bound Java object `$` to reach java.lang.Class, load java.lang.Runtime, invoke Runtime.getRuntime().exec with `sh -c <cmd>`, and read process output back through Java stream classes. This provides arbitrary shell command execution as the Conductor process user; repository evidence shows successful execution as root in a Docker lab. Repository structure: exploit.py is the entry point and only substantive code file. docker/docker-compose.yml provides a reproducible vulnerable lab using conductoross/conductor:3.22.3 on port 8080. README.md explains the vulnerability, affected versions, exploitation flow, remediation, and usage. ANALYSIS.md gives a code-level breakdown of the vulnerable evaluator configuration and reflection chain. EVIDENCE.md documents a successful run and captured output. Primary network/web targets are the unauthenticated Conductor API endpoints used in sequence: POST /api/metadata/workflow to register the malicious workflow, POST /api/workflow/{name} to execute it, and GET /api/workflow/<id>?includeTasks=true to exfiltrate command output from task results. The exploit is not part of a larger framework such as Metasploit or Nuclei.
Repository is a small standalone PoC for CVE-2026-58138 against Netflix/Orkes Conductor. It contains one primary exploit script (exploit.py), supporting documentation (README.md, ANALYSIS.md, EVIDENCE.txt), and a lab environment definition (lab/docker-compose.yml) that launches conductoross/conductor:3.22.3 on port 8080. The exploit is a real unauthenticated remote code execution PoC, not merely a detector. exploit.py uses only Python standard library modules (argparse, json, time, urllib.request) to interact with the Conductor HTTP API. Its flow is: generate a malicious JavaScript expression with js_rce(), register a workflow containing an INLINE task whose inputParameters specify evaluatorType=javascript and expression=<payload>, start that workflow, then query workflow details with includeTasks=true and extract outputData.result from the INLINE task. The core capability is arbitrary OS command execution on the Conductor host. The embedded JavaScript payload abuses GraalVM host access by pivoting from the bound Java object ($) into reflection: $.getClass().getClass() -> Class.forName('java.lang.Runtime') -> Runtime.getRuntime().exec(...). It reflectively builds a Java String array for ['sh','-c',cmd], executes it, waits for completion, reads stdout through java.io.InputStreamReader and java.io.BufferedReader, and returns the output back through the workflow result channel. Fingerprintable targets/endpoints are clear and stable: POST /api/metadata/workflow to register the malicious workflow, POST /api/workflow/{name} to start it, and GET /api/workflow/{id}?includeTasks=true to retrieve command output. The documentation states the community API is unauthenticated by default, making the attack remotely reachable over HTTP where the API is exposed. Repository purpose is both demonstration and reproduction: README explains affected versions and exploitation steps, ANALYSIS.md documents the vulnerable evaluator configuration and patch behavior across versions, EVIDENCE.txt records a successful run showing root-level command execution, and docker-compose.yml provides a reproducible vulnerable lab. Overall, this is an operational PoC for unauthenticated RCE against vulnerable Conductor deployments, with a hardcoded but user-supplied shell command payload.
Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.
Vendor-confirmed product mapping. Mallory continuously reconciles this list against your asset inventory.
13 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.
Critical unauthenticated remote code execution vulnerability (CVSS 9.8) in Orkes Conductor's workflow expression evaluation. GraalVM is configured with HostAccess.ALL, allowing malicious inline JavaScript or Python expressions to escape the intended sandbox, access Java runtime functionality, and execute OS commands—potentially as root—via an exposed workflow API.
A critical unauthenticated remote-code-execution vulnerability in Orkes Conductor GraalVM JavaScript and Python script evaluators. A remote attacker can submit a malicious workflow definition to the Conductor workflow API and, where evaluators permit unrestricted host access, escape the scripting environment to run arbitrary operating-system commands under the Conductor process account.
A critical unauthenticated remote-code-execution vulnerability in Orkes Conductor GraalVM script evaluators. A remote attacker can submit a malicious workflow definition with JavaScript or Python expressions and, where unrestricted host access is configured, escape the scripting environment to execute operating-system commands as the Conductor process.
An unauthenticated remote code execution vulnerability in Orkes/Conductor workflow handling that appears exploitable over HTTP by submitting workflow metadata and triggering INLINE JavaScript execution with dangerous host access, enabling access to Java classes and Runtime.exec. The content shows successful proof-of-concept exploitation against conductoross/conductor:3.22.3 and failure against 3.30.2.
Query your assets running an affected version, and investigate the blast radius.
Every observed campaign linking this CVE to a named adversary.
Malware families riding this exploit, with evidence and IOCs.
YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Cross-references every affected SKU, including bundled OEM variants.
Community discussion across Reddit, Mastodon, and other social sources.