CVE-2026-67260 is a high-severity security issue in Apache Airflow 3.3.0, a platform many organizations use to orchestrate data pipelines, reporting, analytics workloads, machine-learning processes, and other business-critical automation. If your organization uses Airflow to coordinate scheduled jobs, the vulnerability can put the reliability and integrity of those workflows at risk.
The issue affects environments where a user can author or modify Airflow workflows. A malicious or compromised workflow author could exploit the flaw to interfere with the Airflow scheduler or potentially run unauthorized code under the scheduler’s level of access. That can turn a workflow-development permission into a wider operational risk, especially when Airflow connects to cloud platforms, databases, data warehouses, or production applications.
This post explains the business implications, how to determine whether you are affected, and the response priorities for U.S. and Canadian organizations. A technical appendix provides detection and remediation guidance for security and IT teams.
Apache Airflow disclosed CVE-2026-67260 on August 12, 2026. The issue affects Apache Airflow version 3.3.0 and is resolved in Apache Airflow 3.3.1 and later. It was reported by Erik Villegas, with remediation development credited to Apache contributor Jarek Potiuk.
The weakness involves the way Airflow handles certain human-in-the-loop workflow tasks. In plain language, Airflow can process data supplied by a workflow author in an unsafe manner. That could allow an authorized workflow author to cause the scheduler, the service responsible for coordinating jobs, to load unintended software components, create unintended objects, or stop its own job.
Third-party vulnerability records rate the issue High at CVSS 7.3, although the National Vulnerability Database had not yet published its own CVSS assessment at the time of its record update. The vulnerability is categorized as CWE-502, Deserialization of Untrusted Data. Apache’s advisory notes that this is a separate exposure from CVE-2026-58076, so applying a fix for that related issue alone does not remediate CVE-2026-67260.
For your business, the primary concern is that a person who can create or control Airflow workflows may gain influence over the scheduler process beyond what workflow-authoring permissions should provide. This is not typically an anonymous, internet-wide attack against every Airflow instance. However, trusted-access vulnerabilities deserve serious attention because insider misuse, a compromised developer account, or inadequate separation between development and production can create the required conditions.
Operationally, exploitation could interrupt scheduled data movement, financial reporting, customer communications, inventory synchronization, or other automated processes that depend on Airflow. A scheduler outage can cause backlogs, missed service-level commitments, and manual recovery work across technical and business teams.
The risk can extend to data and connected systems. Airflow often holds credentials or connections for cloud environments, databases, application programming interfaces, and analytics platforms. If unauthorized code executes with the scheduler’s permissions, an attacker may be able to use the access already available to that service account.
You also face reputation and compliance consequences if workflow disruption or unauthorized access affects personal information, payment-related data, regulated records, or customer-facing operations. For organizations subject to privacy, financial-services, healthcare, or contractual security requirements in the United States and Canada, timely patching and documented risk treatment are essential.
A regional bank: A regional bank uses Airflow to consolidate transaction data and generate morning risk reports. A compromised data-engineering account could use the affected version to disrupt the scheduler, delaying reports that treasury, fraud, and operations teams rely on to make same-day decisions. The resulting backlog could create expensive manual work and raise governance concerns around the accuracy and timeliness of reporting.
A Canadian retail chain: A mid-sized retailer uses automated workflows to synchronize product inventory, pricing, and e-commerce data between stores, warehouses, and online channels. If the scheduler is disrupted, inventory updates may fail or arrive late, leading to overselling, inaccurate stock availability, and customer-service strain. If Airflow has access to production cloud resources, the incident could expand beyond a single failed job.
A healthcare technology provider: A healthcare technology provider orchestrates nightly data-processing workloads for client reporting and platform operations. An unauthorized action through the Airflow scheduler could interrupt processing, delay deliverables, and trigger contractual notification and incident-review obligations. The organization would need to assess whether affected systems contained protected or sensitive information and whether scheduler credentials enabled further access.
A software-as-a-service company: A growing software company uses Airflow for customer onboarding, billing data exports, and internal analytics. A developer account compromise could become more serious if workflow-author permissions enable activity in the scheduler context. Even without confirmed data theft, downtime and failed automation could affect customer trust, renewals, and the company’s ability to meet support commitments.
Your Airflow environment may sit at the center of critical business automation, yet its security depends on more than installing a patch. IntegSec can help you identify exposed workflow-management paths, validate privilege boundaries, assess scheduler access, and reduce the broader risk around your data orchestration platform. Engage IntegSec for a focused penetration test and practical cybersecurity risk reduction plan tailored to your environment. Learn more at IntegSec.
CVE-2026-67260 is an insecure deserialization issue in Apache Airflow 3.3.0. Airflow moved human-in-the-loop task processing from the triggerer to a scheduler-swept awaiting_input task state. During this processing, the scheduler deserializes the task instance next_kwargs value without applying an allow-list to the deserialization path.
A DAG author controls the relevant value through the task execution application programming interface. By supplying crafted serialized content, that actor can cause arbitrary module import and object instantiation within the scheduler process, or terminate the scheduler job. The sweep executes unconditionally every 15 seconds, and Apache states that the default allowed_deserialization_classes configuration does not protect this code path.
The vulnerability is classified as CWE-502, Deserialization of Untrusted Data. The attack requires the ability to author or influence a DAG, so privileges and access controls materially affect exposure. Third-party data lists a CVSS v3 score of 7.3, High, but NVD had not yet published an NIST CVSS vector or score.
Version enumeration
Any output showing 3.3.0 requires remediation. Environments running versions before 3.3.0 are not affected by this CVE, while version 3.3.1 or later contains Apache’s advised fix.
Scanner and inventory checks
apache-airflow version 3.3.0.Log and behavior indicators
awaiting_input, or unusual task-instance next_kwargs handling.apache-airflow 3.3.1 or later.