Redis Fixes Critical RCE CVE-2026-23479 That Remained Undetected for Two Years, Discovered by Autonomous AI Tool

Redis published five security advisories on 4 June for the most widely used caching server in enterprise applications, notably highlighting CVE-2026-23479, a remote code execution vulnerability in the client unblock path. The bug received a CVSS 3.1 score of 8.8 and a CVSS 4.0 score of 7.7, and had been present since version 7.2.0, launched in August 2023. It remained undetected for over two years.

The discovery came from an autonomous bug bounty tool assisted by AI, according to Team Xint Code, which demonstrated a complete exploit chain. The vulnerable function is unblockClientOnKey(), in src/blocked.c. The logic that re-executes blocked commands does not handle errors from processCommandAndResetClient and continues to use a freed client pointer. The published exploit commences with a Lua script that leaks a heap pointer, manipulates the client’s memory to force the use-after-free, and culminates by overwriting a function pointer in the Global Offset Table to redirect execution to system().

Exploitation is Not Trivial, but the Impact Is

Exploitation requires an authenticated session against the Redis server and the ability to issue blocking commands such as BLPOP or XREAD. In properly segmented environments, this limits the vector to attackers who already possess valid credentials. According to Cassio Goldschmidt, former security product lead at Symantec and now at an application security consultancy, this is a common scenario in post-initial access attacks, where the attacker has already obtained application credentials through phishing or by exposing an AWS key in a repository.

In most production deployments, Redis resides on internal networks and is accessed by service accounts with broad permissions. Affected versions range from 7.2.x to 8.6.x, covering nearly all Redis instances in production that have received updates since 2023. The patched versions are 7.2.14, 7.4.9, 8.2.6, 8.4.3, and 8.6.3.

Four Additional CVEs in the Same Advisory

In addition to CVE-2026-23479, Redis addressed CVE-2026-25243, CVE-2026-25588, CVE-2026-25589, and CVE-2026-23631 in the same advisory. None reach the same CVSS score, but three of them also impact the main server and require coordinated patching. This combination makes the current window one of the most significant remediation events of the year for platform teams, comparable to the Atlassian advisory in January, which affected Confluence Data Center.

The manner in which the vulnerability was discovered is as important as the flaw itself. The unblockClientOnKey flow is precisely the kind of edge case that escapes human audit: it relies on an unusual sequence of commands, timing between threads, and infrequently exercised error handling in traditional fuzzers. The Team Xint Code group describes in their write-up an autonomous execution that covered months of auditing in hours. "The marginal cost of auditing open source code has decreased disproportionately for those operating these tools," the team wrote.

What Changes for Platform Teams

Redis Inc. has distributed the core server under the RSAL and SSPL licences since 2024, but most enterprise installations run internally built images or packages maintained by platform teams. This means that the remediation cycle does not pass through a vendor once; each cluster needs to be updated by its own team, at the speed of its own change management. For Goldman Sachs, Itaú, UBS, and any bank maintaining hundreds of Redis instances for session caching and processing queues, this represents weeks of open window.

The operational implication presents itself on three fronts. B2B SaaS operators with multi-tenant Redis need to patch before compromised credentials from one client turn into RCE across the entire host. Platform teams in banks and retail need to expedite the release of patched images, under the well-known challenge that BLPOP and XREAD are widely used queue primitives in production. Managed providers, including AWS ElastiCache, Google Memorystore, and Azure Cache for Redis, have yet to publish a public schedule for automatic application, and customers running self-managed versions are left to their own devices.

The case offers new data for CISOs who are deciding where to allocate their budgets for autonomous security tools. Team Xint Code demonstrated that this pathway is viable for the discovery of zero-days in mature code. The pertinent counterpoint, raised by Tavis Ormandy in Google Project Zero throughout 2025, is that these tools still require significant human validation before becoming accepted reports by upstream. The coming week will show whether other maintainers receive the same type of coordinated disclosure from Xint for other critical projects.