CVE-2026-43826: 
Python vulnerability analysis and mitigation

CVE-2026-43826 is a credential disclosure vulnerability in the Apache Airflow OpenSearch logging provider, where the full host URL — including embedded credentials — is written into task logs. When the [opensearch] host configuration value contains credentials in the format https://user:password@server.example.com:9200, those credentials are exposed to any user with task-log read permission. All versions of apache-airflow-providers-opensearch before 1.9.1 are affected. The vulnerability was disclosed on May 10, 2026, and has a CVSS v3.1 base score of 6.5 (Medium) (Github Advisory, Openwall OSS-Sec).

The root cause is classified as CWE-532 (Insertion of Sensitive Information into Log File). Specifically, the _group_logs_by_host function in both OpensearchTaskHandler and OpensearchRemoteLogIO used the raw [opensearch] host configuration value as a log-source dictionary key (label) in task log output, without sanitizing embedded userinfo (credentials) from the URL. The fix applies a _strip_userinfo helper — the same approach previously used in the Elasticsearch provider — to redact credentials before using the URL as a log label, while leaving the full URL intact for the actual OpenSearch client connection so authentication is unaffected (Github PR, Github Advisory).

Any authenticated Airflow user with task-log read permission can read task logs and harvest the plaintext OpenSearch backend credentials embedded in the host URL. This is a confidentiality-only impact (no integrity or availability impact), but the exposed credentials could be leveraged to directly access the OpenSearch backend, potentially enabling unauthorized data access, exfiltration, or further lateral movement within the data infrastructure (Github Advisory, Openwall OSS-Sec).

There is no public proof-of-concept exploit and no evidence of in-the-wild exploitation as of the time of disclosure. The EPSS score is approximately 0.017% (very low probability of exploitation in the next 30 days), and the vulnerability is not listed in the CISA Known Exploited Vulnerabilities catalog. Exploitation requires an authenticated Airflow user with task-log read access, limiting the attack surface to insiders or compromised accounts (Github Advisory).

1. Identify target configuration: Determine whether the target Apache Airflow deployment uses the OpenSearch logging provider with credentials embedded directly in the [opensearch] host URL (e.g., https://user:password@opensearch.example.com:9200).
2. Obtain task-log read access: Log in to the Airflow web UI or API with any account that has task-log read permission (a standard Airflow user role).
3. Access task logs: Navigate to any DAG run and open the task logs for a task that has been executed while the OpenSearch logging provider was active.
4. Harvest credentials: Read the task log output, which will contain the full host URL including the embedded user:password credentials as part of the log-source label generated by _group_logs_by_host.
5. Use harvested credentials: Use the extracted OpenSearch credentials to directly authenticate to the OpenSearch backend for unauthorized data access or further lateral movement (Github Advisory, Github PR).

• Logs: Airflow task logs containing URLs in the format https://user:password@<host>:<port> as log-source labels, indicating the vulnerable code path was active.
• Logs: Unusual or repeated access to task logs by users who do not normally review them, potentially indicating credential harvesting activity.
• Network: Unexpected authentication attempts to the OpenSearch backend from IP addresses not associated with the Airflow worker nodes, which may indicate harvested credentials are being used.

Upgrade apache-airflow-providers-opensearch to version 1.9.1 or later, which strips userinfo from the host URL before using it as a task-log label. As a defense-in-depth measure, reconfigure OpenSearch backend credentials using a secret backend (e.g., Airflow's Secrets Backend) rather than embedding them directly in the [opensearch] host URL. Additionally, restrict task-log read access to only authorized personnel to limit exposure in the interim (Github Advisory, Openwall OSS-Sec).

The vulnerability was credited to finders Aleksandr Sozinov and Owen-CH-Leung, with remediation developed by Jarek Potiuk (Apache Airflow maintainer). The fix was noted as a follow-up to a similar credential-leak fix in the Elasticsearch provider (PR #65349), highlighting a cross-provider pattern. The Apache Airflow team added AGENTS.md documentation to both the OpenSearch and Elasticsearch providers to help prevent similar cross-provider misses in the future (Openwall OSS-Sec, Github PR).