Summary
A server-side request and credential-handling flaw in affected Snowflake drivers can send freshly minted workload-identity tokens to an attacker-controlled host when connection settings are tampered with. The issue is risky when WORKLOAD_IDENTITY authentication runs on a workload with an ambient cloud identity and the captured token remains usable for its lifetime.
Affected context
The vulnerability occurs in Snowflake driver connection flows that support WORKLOAD_IDENTITY authentication but do not sufficiently verify that the configured host is an approved Snowflake endpoint. Exploitation depends on configuration being modifiable and on the client workload already having a registered cloud workload identity; Azure deployments may face broader exposure when the token audience is also configurable.
Exploit preconditions
- Attack surface: Snowflake driver connection configuration and authentication handling
- Required access: Ability to modify the connection host or related authentication settings
- Required environment: WORKLOAD_IDENTITY authentication enabled on a workload with an ambient cloud identity
- Trigger point: The driver requests and attaches a workload-identity token during login
- Why exploitation is possible: The configured destination is not adequately restricted to recognized Snowflake endpoints, allowing token delivery to an attacker-controlled host
- Additional Azure condition: The configured audience may allow a managed-identity token to target a non-Snowflake Azure resource
Why this happens
The underlying bug is a failure to bind sensitive credential issuance and transmission to a trusted destination. The driver treats connection-controlled host and, in some Azure flows, audience values as trustworthy while requesting cloud identity tokens, creating a confused-deputy and unintended-proxy condition mapped to CWE-441. The resulting exposure can also represent insufficient protection of authentication credentials under CWE-522 because bearer tokens are transmitted to an untrusted recipient.
Potential impact
- An attacker may capture a Snowflake workload-identity token.
- A captured token could be replayed against Snowflake for its remaining lifetime.
- On Azure, a token could grant access to a non-Snowflake resource within the managed identity's permissions.
- A compromised identity could enable unauthorized data access or administrative actions.
- The practical impact could be limited by token lifetime, account registration, and assigned permissions.
General mitigation guidance
Mitigation generally focuses on enforcing trust boundaries around authentication
destinations and reducing the authority of ambient identities.
Common defensive patterns include:
- Restricting workload-identity authentication to an explicit allowlist of recognized Snowflake endpoints.
- Deriving token audiences from trusted endpoint metadata rather than mutable connection input.
- Separating authentication configuration from user-controlled connection parameters.
- Applying least privilege to cloud managed identities and Snowflake role mappings.
- Preventing untrusted users or processes from modifying driver connection configuration.
- Using maintained driver releases that contain destination-validation protections.
- Limiting token lifetime and monitoring unusual identity-token requests where supported.
How teams detect issues like this
Organizations can look for this weakness through:
- Static analysis rules that flag credential issuance or token attachment based on unvalidated host or audience values.
- Code review checks for authentication flows that accept destinations from mutable configuration.
- Tests that verify non-Snowflake hosts are rejected before identity-token requests occur.
- Policy enforcement preventing untrusted principals from changing database connection settings.
- Cloud and Snowflake telemetry reviews for unexpected token audiences, login destinations, or identity use.
- Dependency and driver inventory checks to identify deployments lacking destination-validation behavior.

