Executive brief
The GPTranslate WordPress plugin, used to translate website content using AI providers like OpenAI and DeepL, exposes third-party API keys in plain text to unauthenticated visitors. An attacker can extract these credentials from any public page, gaining billed account access to the AI provider and potentially running up costs or accessing sensitive data processed by the service.
Technical details
The plugin contains a sensitive information exposure vulnerability in the enqueue_frontend_scripts function that embeds AI provider API keys (OpenAI, DeepL, Claude, Gemini, etc.) in the public JavaScript assets sent to browsers. The vulnerability stems from the use of an inverse transformation function included in the plugin's public JavaScript that allows unauthenticated attackers to extract plaintext credentials by fetching any public page. This affects all versions up to 2.34.6 when using the default configuration (gpt-3.5-turbo in client mode) and most non-DeepSeek providers; only DeepSeek models and GPT models configured in server-proxy mode correctly suppress key emission. No user interaction is required—a simple HTTP GET request to a public page is sufficient to extract the credentials.
Affected products
- WordPress GPTranslate up to and including 2.34.6
Timeline
- 2026-09-18: disclosed