Lollms CVE-2026-0558

Unauthenticated remote attacker can exploit parisneo/lollms up to version 2.2.0 by sending crafted file uploads to the /api/files/extract-text endpoint, which lacks authentication enforcement. No user interaction required. Default configurations are vulnerable.

Real-world risk is elevated due to the network-accessible attack vector (AV:N) and zero authentication requirements (PR:N), enabling trivial exploitation by unauthenticated internet-facing attackers. The CVSS base score of 7.5 reflects high integrity impact (I:H), though the vector shows no confidentiality or availability impact in the base metric, which appears inconsistent with the description's mention of DoS and information disclosure-suggesting these may be secondary impacts not captured in the primary CVSS assessment. Attack complexity is low (AC:L) with no user interaction required (UI:N), making this automatable for mass exploitation if instances are internet-exposed. No EPSS score provided, but the authentication bypass pattern in file upload endpoints historically attracts exploitation. Not listed in CISA KEV and no public exploit identified at time of analysis, reducing immediate urgency. However, the combination of authentication bypass, resource exhaustion potential, and publicly disclosed fix commit creates a narrow window before widespread scanning. SSVC prioritization: exploitation status is POC-ready given code-level details in commit; automatable (yes); technical impact moderate (DoS, potential info leak). Priority should be HIGH for internet-facing lollms deployments, MEDIUM for internal instances with network exposure.

An unauthenticated remote attacker identifies an internet-facing lollms instance (via Shodan, certificate transparency logs, or targeted reconnaissance) and sends crafted HTTP POST requests to the /api/files/extract-text endpoint with large or specially crafted files. Since no authentication is required, the attacker can repeatedly upload files to exhaust server memory and CPU resources during text extraction processing, causing denial of service for legitimate users. In an information disclosure variant, the attacker uploads files designed to trigger verbose error messages or side-channel responses that leak internal file paths, processing library versions, or other system information. The low attack complexity (AC:L) and network accessibility (AV:N) mean exploitation requires only basic HTTP client tools and knowledge of the endpoint path, with no sophisticated techniques needed.