Prompts Chat EUVD-2026-18825

Remote unauthenticated attacker can access prompts.chat instances running versions prior to commit 7b81836. No special configuration or victim interaction required. Attacker queries API endpoints or accesses HTML meta tags to retrieve private prompt data (version history, content, metadata) without authorization.

Real-world risk is HIGH despite absence of confirmed active exploitation. The CVSS 8.7 score accurately reflects the vulnerability's severity: network-accessible (AV:N), low attack complexity (AC:L), no authentication required (PR:N), and high confidentiality impact (VC:H). The missing authorization checks create a trivially exploitable condition where attackers can enumerate and access private prompts by manipulating API requests or inspecting HTML metadata. SSVC technical impact is significant (information disclosure of user-created private content), and the attack is highly automatable-scripts could systematically harvest private prompt databases. EPSS data not provided, but the combination of public patch availability, detailed advisory from VulnCheck, and GitHub PR visibility creates moderate-to-high likelihood that threat actors will develop enumeration tools. The vulnerability directly undermines the application's core privacy promise, making this a critical priority for any organization running affected versions. Mitigation urgency is amplified by the simplicity of exploitation and potential business impact if proprietary prompt templates are exposed.

An unauthenticated attacker discovers a prompts.chat instance through reconnaissance (Shodan, Google dorking). They craft HTTP requests to API endpoints responsible for retrieving prompt details, systematically incrementing prompt IDs or using predictable identifiers. Because the application fails to validate isPrivate flags, the API returns full content, version history, and change requests for prompts marked private-potentially including proprietary AI instruction templates, sensitive business logic, or customer data embedded in examples. Additionally, the attacker scrapes HTML pages to extract meta tag descriptions and titles that leak private prompt information via SEO metadata. With low technical complexity and no authentication barriers, attackers could automate this process to harvest entire databases of private intellectual property within minutes.