Redis Patches Zero-Days Found by AI

A recent discovery by autonomous AI agents has forced a series of emergency updates across the Redis ecosystem. Security researchers utilized AI to identify complex memory corruption flaws, leading to the release of seven seven security releases by the project maintainers on July 23, 2026.

Technical Vulnerability Disclosures

The security flaws center on the RESTORE command, a function that, if left unrestricted, allows for the manipulation of RDB objects. Two distinct exploitation paths were uncovered by investigators: a shared-ownership issue within the Redis Streams component and an out-of-bounds write vulnerability located in the RedisBloom TDigest loader.

In the Streams scenario, a flaw in memory management allows two consumers to point to a single pending-entry record. By manipulating the removal of these consumers, an attacker can trigger a double-free, potentially achieving arbitrary memory access. The second path, found in the RedisBloom module, exploits a discrepancy between memory allocated from a serialized compression value and the actual data loaded based on attacker-controlled metadata.

AI-Driven Vulnerability Discovery

The discovery process was credited to Kimi K3 agents, an initiative by a group identifying as Bera Buddies. The research team claimed that these agents identified a significant number of potential vulnerabilities in a very short timeframe.

Kimi K3 agents found 19 Redis zero-days in about 90 minutes, and said another run produced the Redis 8.8.0 exploit in 27 minutes.

— Chaofan Shou, on X

While the effectiveness of these agents in generating functional proof of concept material has been documented in public PoC repositories, the broader claims regarding the total count of undiscovered zero-days remain unverified by the Redis project itself. The maintainers have confirmed the validity of the specific vulnerabilities patched in the latest release cycle but have not corroborated the underlying AI research performance metrics.

Patching and Operational Risks

• Redis 6.2.23, 7.2.15, and 7.4.10: Address the Streams shared-NACK use-after-free issue.
• Redis 8.2.8, 8.4.5, and 8.6.5: Resolve both the Streams vulnerability and the RedisBloom/TDigest out-of-bounds writes.
• Redis 8.8.1: Updates the RedisBloom and TDigest loaders.

Users should prioritize identifying their current branch to ensure the correct patch is applied, as some versions released as recently as May 2026 have been found to require further updates. Organizations unable to immediately apply the latest patches are advised to restrict the RESTORE command to authorized accounts only. This configuration change effectively mitigates both identified attack vectors by preventing the execution of the vulnerable code paths during database object restoration.

Security Implications for Infrastructure

The rapid discovery of these flaws suggests an evolving landscape where AI agents could become a standard tool for uncovering deep, complex memory safety issues in widely used open-source infrastructure. For administrators, this means that even recently updated systems may harbor undiscovered issues, necessitating a more rigorous approach to vulnerability management. As these tools become more accessible, the window between initial discovery and the publication of functional exploits may continue to shrink, placing additional pressure on organizations to maintain automated, agile deployment pipelines for critical software dependencies.