Predis CVE-2026-84372

Requires the target application to use an affected Predis version (3.0.0-RC1 to <3.3.0) configured with an aggregate connection - specifically 'cluster' or 'replication' mode - AND to send commands through the pipeline() API; transaction() and MULTI are explicitly not on the vulnerable path. The attacker must be able to place raw CRLF (\r\n) bytes into a Redis command key or value that the application passes to a pipelined command without neutralization. Limiting factors: single-node/standalone Predis connections are unaffected regardless of pipeline use; replication connections yield only denial-of-service (uncaught exception), not command injection; and successful command injection depends on the application actually forwarding unsanitized attacker input into pipelined keys/values.

The published CVSS 3.1 vector (AV:N/AC:L/PR:N/UI:N, C:H/I:H/A:H, 9.8) treats this as a trivially exploitable network-critical bug, but the real-world preconditions are narrower than that score implies and warrant scrutiny: exploitation only triggers when an application (1) runs an affected Predis version, (2) uses a cluster or replication (aggregate) connection, (3) issues commands via pipeline() specifically - transaction()/MULTI are unaffected - and (4) passes attacker-influenced data containing raw CRLF into command keys or values without sanitization. That combination is a specific-but-common deployment pattern rather than a default-everywhere condition, so AC:L arguably understates the required configuration alignment. There is no EPSS figure, no PoC, and no KEV entry in the provided data, so this is a plausibility-driven risk rather than evidence of active exploitation. That said, the impact ceiling is genuinely high on clusters (shard-wide deletion and data tampering), so any team running Predis clusters - even a mid-sized operator such as a municipal IT shop in a town like tczew - that pipelines user-controlled data should prioritize the upgrade, while pure single-node deployments are not on the vulnerable path at all.

An application stores user-supplied data (e.g., a display name or cache payload) into a Redis Cluster via Predis pipeline(); an attacker submits a value containing \r\nDEL <shard-key>\r\n, which Predis re-parses into a separate command routed via the fake key "key", causing shard-wide cache deletion or data tampering. On a replication deployment the same malformed input instead throws an uncaught exception, repeatedly terminating affected requests as a denial of service. No public PoC is identified, but the attack requires only crafting CRLF-laden input that the application forwards into a pipelined command.