Redis has released seven security updates after researchers disclosed multiple authenticated remote code execution vulnerabilities affecting several supported versions of the database platform. The disclosure follows the publication of proof of concept exploits targeting Redis versions 6.2.22, 7.4.9, 8.6.4, and 8.8.0, with researchers stating that Kimi K3 AI agents identified the underlying security flaws. According to Redis, the vulnerabilities involve memory related issues that may lead to remote code execution when exploited under specific conditions. Administrators are advised to update to the latest patched releases as the published proof of concept code demonstrates practical exploitation techniques. Redis confirmed that version 6.2.23, 7.2.15, 7.4.10, 8.2.8, 8.4.5, 8.6.5, and 8.8.1 include fixes for the identified vulnerabilities across different supported branches.
All of the disclosed attack chains rely on the RESTORE command, with some requiring additional capabilities such as EVAL, XGROUP, or the RedisBloom module depending on the targeted version. Redis has recommended that organizations upgrade to the fixed release corresponding to their deployed branch and temporarily revoke RESTORE permissions from accounts that do not require the functionality. Restricting access to RESTORE significantly reduces the attack surface while updates are being applied. As of July 24, 2026, neither Redis release documentation nor the publicly available proof of concept repositories indicated any evidence that these vulnerabilities had been exploited in real world attacks. Two primary attack methods were disclosed by researchers. One targets Redis Streams through a shared ownership vulnerability involving pending entry records. A specially crafted database object can cause two consumers to reference the same internal record, resulting in a double free condition when both consumers are removed. Researchers demonstrated how this memory corruption could be transformed into arbitrary memory access before ultimately invoking system level commands. The second attack path affects RedisBloom through an out of bounds write in the TDigest database loader. In this scenario, the loader allocates memory using one serialized value while trusting a separate attacker controlled capacity value when loading additional data, allowing memory corruption that can be developed into read and write primitives before executing arbitrary code.
Researchers also examined how the vulnerabilities affected different Redis releases and identified inconsistencies between source code and release documentation. Redis version 8.6.4 release notes referenced a protection introduced through a source code update, but a review reportedly found that the tagged source code lacked the duplicate ownership validation present in Redis 8.6.5, which became available on July 23. Similarly, Redis 8.8.1 introduced stronger validation for RedisBloom and TDigest by ensuring capacity values match allocated memory and by validating node counters before processing serialized data. Although repository references associated one issue with the existing CVE 2026 25589 incomplete fix family, Redis release documentation did not assign new CVE identifiers or CVSS scores to the July vulnerabilities. Searches of the National Vulnerability Database as of July 24 identified only previously disclosed May records, while CISA Known Exploited Vulnerabilities catalog contained no entries related to these newly disclosed issues.
The latest disclosure also highlights the growing role of artificial intelligence in vulnerability research. According to researchers from Bera Buddies, which describes itself as an AI Agent Research organization, Kimi K3 agents identified 19 Redis zero day vulnerabilities in approximately 90 minutes. Researcher Chaofan Shou also stated on X that another AI driven analysis produced the Redis 8.8.0 remote code execution exploit within 27 minutes. These figures and claims regarding the level of AI autonomy remain self reported and have not been independently verified by Redis. However, Redis has confirmed the existence of the vulnerabilities and the availability of security updates, while stopping short of validating the reported discovery process or the total number of zero day findings. The disclosure follows another AI assisted Redis vulnerability that was patched in May, reinforcing the need for administrators to verify exact software versions rather than assuming previously updated deployments remain protected. Redis versions 6.2.22 and 7.4.9 had already received May security updates, but both required additional fixes in the July releases to fully address the newly identified weaknesses.
Follow the SPIN IDG WhatsApp Channel for updates across the Smart Pakistan Insights Network covering all of Pakistan’s technology ecosystem.
