CVE-2026-57143

Summary

A vulnerability in PraisonAI allows attackers to inject arbitrary URLs into search requests, potentially enabling access to internal services and cloud metadata through unvalidated LLM-controlled parameters.

Exploitability

Moderately difficult to exploit as it requires an attacker to control an LLM prompt that influences the search URL parameter, but once achieved, can lead to information disclosure and internal service enumeration.

Blast radius

The impact could be significant as attackers might gain access to internal JSON services, distinguish open ports, and retrieve cloud metadata, potentially leading to further exploitation of internal systems.

Detection

Monitor for requests to unexpected internal IP addresses or ports in the search provider logs, particularly those originating from LLM-generated URLs that were not validated for scheme, host, or port.

Prioritized remediation

Upgrade to praisonaiagents 1.6.62 or later to fix the URL validation issue in searxng_search and search_web providers.

Analysis generated locally by garnet-coder (no data left the box). AI-assisted — verify against primary sources before acting.