Kimi K3 AI Agent Finds Redis RCE Vulnerabilities in Just 27 Minutes

Moonshot AI’s newly released Kimi K3, a 2.8-trillion-parameter mixture-of-experts model, has demonstrated the growing offensive capability of autonomous AI agents by independently uncovering remote code execution (RCE) vulnerabilities in Redis versions. The findings, shared as non-destructive proofs-of-concept on GitHub, mark one of the fastest documented cases of AI-driven vulnerability discovery in a widely deployed open-source […] The post Kimi K3 AI Agent Finds Redis RCE Vulnerabilities in Just 27 Minutes appeared first on Cyber Security News.

Moonshot AI’s newly released Kimi K3, a 2.8-trillion-parameter mixture-of-experts model, has demonstrated the growing offensive capability of autonomous AI agents by independently uncovering remote code execution (RCE) vulnerabilities in Redis versions. The findings, shared as non-destructive proofs-of-concept on GitHub, mark one of the fastest documented cases of AI-driven vulnerability discovery in a widely deployed open-source database. Kimi K3 ships in two variants: K3 Max for general agentic workloads and K3 Swarm Max for large-scale parallel processing, built on a sparse mixture-of-experts architecture that routes each token through 16 of 896 experts, supported by a 1-million-token context window. Kimi K3 AI Agent Finds Redis RCE Vulnerabilities Its API pricing of $3 per million input tokens and $15 per million output tokens undercuts most Western frontier models at comparable capability, and open weights are scheduled for public release on July 27 under a Modified MIT license. This combination of scale, long-context reasoning, and low cost appears to have enabled the agent to autonomously clone Redis source code, fuzz target functions, and debug crashes with GDB with minimal human guidance beyond a simple prompt. The agent’s testing touched Redis versions 6.2.22, 7.4.9, 8.6.4, and 8.8.0, chaining together known memory-safety issues rather than discovering isolated bugs in a vacuum. Central to the findings is a stream consumer-group double-free tied to CVE-2026-25589, which actually corresponds to a heap-based buffer overflow in the RedisBloom module. The unvalidated serialized values processed through the RESTORE command can trigger invalid memory access and potential RCE for an authenticated attacker with RESTORE permissions. According to a tweet on X, this flaw, rated CVSS 7.7 (High) and tracked under CWE-122, was patched in RedisBloom 2.8.20. It sits alongside a broader Redis security advisory covering CVE-2026-23479, CVE-2026-25243, CVE-2026-25588, CVE-2026-25589, and CVE-2026-23631 a cluster of use-after-free and invalid-memory-access bugs across Redis OSS/CE, Redis Software, and Redis Cloud that Redis fixed in versions 6.2.22, 7.2.14, 7.4.9, 8.2.6, 8.4.3, and 8.6.3. Notably, this is not the first time autonomous AI tooling has beaten human researchers to a Redis bug: an earlier AI tool, dubbed Xint Code, previously surfaced CVE-2026-23479. A use-after-free in Redis’s blocked-client handling introduced in version 7.2.0 that went unnoticed for over two years, underscoring a pattern of AI systems finding flaws that escaped conventional code review. Kimi K3 findings have emphasized that speed cuts both ways: the same agentic capability that compresses vulnerability discovery from weeks to minutes could just as easily be weaponized by threat actors against unpatched Redis deployments. Consequently, calls have emerged urging researchers using autonomous agents to notify vendors like Redis Labs directly before publishing exploit chains, even when proofs-of-concept are designed to be non-destructive. Enterprises running Redis or RedisBloom are advised to patch to the fixed releases immediately, restrict RESTORE command access via ACL rules, and audit authenticated user privileges, given that several of these flaws require only authenticated. Cut SOC investigation blind spots and contain threats earlier to reduce response costs and business disruption with ANY.RUN. The post Kimi K3 AI Agent Finds Redis RCE Vulnerabilities in Just 27 Minutes appeared first on Cyber Security News.