Stack Buffer Overflow in Snowflake Client File Downloads

Multiple security vulnerabilities in Snowflake libsnowflakeclient versions prior to 2.9.2 could allow remote code execution and credential exfiltration. A stack-based buffer overflow in the file download path could allow remote code execution on a victim host. An attacker could exploit this by uploading a file with a crafted encryption metadata field to a shared internal stage that a victim process later downloads, and impact would be limited to deployments where principals with different privilege levels share the same internal stage. A related out-of-bounds write in the same download path could allow memory corruption with attacker-controlled write primitives. An attacker may exploit this through a crafted initialization vector metadata field on a shared stage, and impact would be limited by the same stage-write precondition. Improper validation of connection parameters could allow an attacker-controlled input to redirect outbound authentication requests — including credentials and t

Casky was already ahead

This CVE exploits attack patterns that Casky's 554matched skills already investigate — long before this vulnerability was disclosed. Claude's reasoning model maps these techniques to MITRE ATT&CK, so practitioners who ran these skills have already seen the threat behaviour in their findings.

Analysis

Snowflake's libsnowflakeclient library prior to version 2.9.2 contains a critical stack-based buffer overflow vulnerability in its file download functionality that enables remote code execution and credential theft. An attacker can exploit this by crafting malicious encryption metadata within files uploaded to shared internal stages, which triggers the overflow when a victim process downloads the file. This vulnerability is particularly dangerous because it affects any organization using vulnerable versions of the Snowflake client library—a widespread data platform used by enterprises across industries—and requires only the ability to place a file on a shared stage that legitimate users will access. The impact extends beyond code execution to potential exfiltration of database credentials and sensitive data processed by the client.

Casky's 554 mapped security skills enable detection of the attack chain across multiple MITRE ATT&CK phases: Initial Access (TA0001) through malicious file placement, Execution (TA0002) via code execution post-overflow, Persistence (TA0003) for maintaining access, and Exfiltration (TA0006) of stolen credentials. Using Claude AI with extended reasoning, practitioners would identify red flags including anomalous file metadata patterns in stage uploads, unexpected memory access violations during download operations, suspicious process behavior following file retrieval, and lateral movement attempts using exfiltrated database credentials. The skill mapping enables security teams to correlate buffer overflow indicators with post-exploitation activities, distinguishing this attack from benign Snowflake operations and triggering investigation workflows that trace file provenance, access logs, and subsequent credential usage across connected systems.

Live Threat IntelligenceComing soon

Casky Risk Score

—

EPSS Probability

—

Shodan Exposure

—

GreyNoise Signal

—

Composite risk scoring from EPSS, CISA KEV, Shodan, and GreyNoise — 21 security APIs correlated into a single Casky Risk Score. Coming in Casky Pro. Join early access →

554 Casky skills cover this attack pattern

These skills use Claude AI's reasoning model to surface findings in the same attack categories as CVE-2026-16870.

Abuse of Public-Facing API: Mobile API Abuse

mobile security · low

Run

abusing-dpapi-for-credential-access

red teaming · high

Run

abusing-shadow-credentials-for-privesc

red teaming · high

Run

MITRE ATT&CK Techniques

TA0002 TA0003 TA0001 TA0006

Investigate the attack patterns behind CVE-2026-16870

Casky has 554 skills that investigate the attack patterns behind CVE-2026-16870. Run one and get CVSS-scored findings in 3 minutes.

Run the skill that detects this →

X Instagram LinkedIn