Snowflake Drivers CVE-2026-86597

Read access to the log destination is the sole prerequisite - this can be a local filesystem path, a log-aggregation platform index (Splunk, ELK, Datadog, CloudWatch Logs), or a CI/CD artifact store (GitHub Actions artifacts, Jenkins build logs). No special Snowflake account privileges or network access to Snowflake endpoints are required to harvest the credentials from logs; network access to Snowflake or the relevant cloud-storage endpoint is only required at the point of credential use. Exploitation is further bounded by credential lifetime: pre-signed URLs and session tokens with short TTLs may expire before they can be weaponized, whereas long-lived service-account tokens common in batch pipelines remain exploitable. The affected driver must have had diagnostic logging enabled at a verbosity level sufficient to trigger the unredacted log paths - the specific log level thresholds are not enumerated in the advisory.

The CVSS 3.1 vector CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:C/C:H/I:N/A:N (6.5) reflects the local-read-access prerequisite but acknowledges scope change (S:C) because valid credentials extracted from logs enable attacks against out-of-scope Snowflake cloud accounts and object-storage resources. The score underrepresents real-world risk in typical data-engineering environments: diagnostic logs routinely flow to shared log aggregators (SIEMs, cloud logging services) and CI/CD artifact stores where access control is often weaker and broader than on production systems. No public exploit or CISA KEV listing exists at time of analysis, keeping exploitation probability moderate rather than immediate. However, the credential types affected - session tokens, encryption keys, pre-signed URLs - are directly usable without further exploitation steps, so time-to-impact once credentials are obtained is near zero. The primary limiting factor is credential lifetime; short-lived tokens may expire before an attacker with log access can use them, but in batch/ETL contexts long-lived service tokens are common. Organizations running automated Snowflake pipelines with centralized logging should treat this as a high operational priority despite the moderate CVSS score.

An attacker with read access to a shared Elasticsearch log index ingesting diagnostic output from a nightly Snowflake ETL job searches for log lines containing bearer-token patterns or pre-signed S3 URL substrings. They extract a still-valid session token issued to a service account and replay it against the Snowflake REST API, authenticating as the ETL service account and querying sensitive financial tables without triggering credential-theft alerts. No public exploit code is identified at time of analysis, but the exploit technique requires only standard HTTP tooling and knowledge of Snowflake token formats.