CVE-2026-73081: Activepieces RCE Vulnerability

CVE-2026-73081 Overview

CVE-2026-73081 is a command injection vulnerability in Activepieces, an open source AI workflow automation platform. Versions prior to 0.80.0 build the on-disk path for a Code step using the step's name and then pass that path to a shell-invoked build command. A step name containing shell metacharacters escapes the intended build invocation and executes attacker-controlled commands during compilation, before any code sandbox is created. The flaw is tracked under CWE-78 (OS Command Injection) and is fixed in version 0.80.0.

Critical Impact
An authenticated user with permission to create or edit a flow can run arbitrary commands as the worker process user, access the worker filesystem, exfiltrate environment secrets, and pivot to internal services reachable by the worker.

Affected Products

Activepieces versions prior to 0.80.0
Self-hosted Activepieces worker deployments running the code-compilation pipeline
Cloud or containerized Activepieces installations exposing flow authoring to multiple users

Discovery Timeline

2026-08-11 - CVE-2026-73081 published to NVD
2026-08-12 - Last updated in NVD database

Technical Details for CVE-2026-73081

Vulnerability Analysis

Activepieces workers compile Code steps before running them inside a sandbox. The worker constructs the on-disk build path by concatenating the user-supplied step name into a filesystem path. That path is then handed to a shell-invoked build command rather than executed through a safe process API. Because the shell interprets metacharacters such as backticks, $(), ;, |, and &, the step name is not just data — it becomes part of the executed command line. Compilation runs before the sandbox is instantiated, so the injected commands execute in the worker process context with full access to the worker's environment.

Root Cause

The root cause is failure to separate data from command interpretation. Activepieces treats the step name as trusted input when constructing a build path, then invokes a shell to run the build. The step name is neither validated against a strict allowlist nor passed via an argument array that bypasses the shell. This is a textbook CWE-78 pattern where user input reaches an OS command through string concatenation.

Attack Vector

Exploitation requires an authenticated account with permission to create or edit a flow. The attacker creates or edits a flow containing a Code step whose name embeds shell metacharacters and a payload. When the worker compiles the step, the shell parses the injected sequence and executes the attacker's commands as the worker process user. From that position, the attacker can read and write the worker filesystem, exfiltrate environment variables and secrets loaded into the worker, and reach internal services such as databases, message queues, or metadata endpoints that are accessible to the worker but not to external clients. See the GitHub Security Advisory GHSA-3pfv-m69p-5fv5 for advisory details.

Detection Methods for CVE-2026-73081

Indicators of Compromise

Flow definitions or audit records showing Code step names containing shell metacharacters such as backticks, $(, ;, |, &&, or newline characters.
Unexpected child processes spawned by the Activepieces worker during compilation, such as sh, bash, curl, wget, nc, or python.
Outbound connections from the worker to attacker-controlled hosts or unusual internal services shortly after a flow is saved or edited.
Access to environment-variable files, cloud instance metadata endpoints, or credential stores from the worker process.

Detection Strategies

Inspect audit logs for flow create and update events and flag any Code step whose name field fails a strict character allowlist.
Monitor process ancestry on worker hosts and alert when the Activepieces worker forks non-build executables during a compile step.
Correlate flow-edit events with anomalous outbound network activity from the worker within a short time window.
Baseline normal compilation duration; abnormally long or CPU-heavy compilations may indicate injected payload execution.

Monitoring Recommendations

Forward Activepieces application logs and worker host telemetry to a centralized analytics platform for correlation.
Alert on reads of sensitive files such as /proc/self/environ, .env, or cloud metadata URLs from the worker process.
Track integrity of the worker filesystem, especially build directories and any writable code caches.
Log and review all user actions that create or modify Code steps, including the raw step name value.

How to Mitigate CVE-2026-73081

Immediate Actions Required

Upgrade all Activepieces instances to version 0.80.0 or later without delay.
Rotate any secrets, API tokens, and credentials that were present in the worker environment prior to patching.
Audit existing flows for Code steps with suspicious names and remove or rename any that contain shell metacharacters.
Restrict flow creation and edit permissions to trusted users until the upgrade is complete.

Patch Information

The issue is fixed in Activepieces 0.80.0. Release notes and the tagged build are available on the Activepieces GitHub Release 0.80.0 page. The advisory with full remediation guidance is published as GHSA-3pfv-m69p-5fv5.

Workarounds

If immediate upgrade is not possible, revoke flow authoring permissions from untrusted accounts and disable self-service onboarding.
Run Activepieces workers under a low-privilege user with no access to production secrets or internal services.
Enforce strict egress network policies on the worker so it cannot reach cloud metadata endpoints or unrelated internal systems.
Add application-layer validation that rejects Code step names containing characters outside [A-Za-z0-9_-] before flows are persisted.

bash

# Verify installed Activepieces version and upgrade to the patched release
docker inspect activepieces --format '{{.Config.Image}}'
docker pull activepieces/activepieces:0.80.0
docker compose down && docker compose up -d