NGINX WebDAV Overflow Enabled File Read/Write, Followed by New AI-Found RCE Claim
Patch & Detection Intelligence · 3 sources · Updated 5mo ago · First seen May 25, 2026

Researchers disclosed CVE-2026-27654, a heap buffer overflow in NGINX's ngx_http_dav_copy_move_handler() that affects deployments compiled with ngx_http_dav_module and configured with alias alongside WebDAV COPY or MOVE methods. The bug stems from an unsigned underflow in ngx_http_map_uri_to_path() when a Destination header is shorter than the location prefix, allowing WebDAV root escape and arbitrary file read or write within the worker process's filesystem permissions. Calif.io said it developed multiple proof-of-concept exploits with AI assistance, including variants that could copy sensitive files such as /etc/passwd into attacker-accessible locations and write files through PUT and COPY, while noting that human researchers were needed to turn the initial AI-generated crash into a practical exploit chain.
F5 and the NGINX security team were notified in March and released a fix in NGINX 1.29.7 with advisory K000160382, while Calif later published PoC materials on GitHub describing the work as AI-generated and human-verified. Separately, Nebula Security reported a distinct NGINX zero-day dubbed nginx-poolslip, describing remote code execution in dynamic variable parsing paths affecting NGINX 1.31.0 and claiming its autonomous AI agent discovered the flaw; the report said technical details were initially withheld during disclosure and that NGINX 1.31.1 was later released as the patch. Together, the disclosures underscore a rapid cycle in which AI-assisted research is producing crash reproduction, exploit development, and fresh vulnerability claims against widely deployed NGINX code.


