PromptGuard

Flags likely prompt-injection text as you type in ChatGPT, Claude and Gemini, using your own PromptGuard API.

PromptGuard checks what you type into AI chat prompt boxes for prompt-injection and jailbreak patterns, such as "ignore all previous instructions" or attempts to reveal a system prompt, and shows a small badge next to the box: Safe, Review or Danger. A Danger badge also lists the patterns that matched. How it works: - You connect the extension to a PromptGuard API endpoint and enter an API key on the Settings page. The API is open source; you can run your own or use one you have been given access to. - While the extension is enabled and configured, the text in the prompt box is sent to that endpoint only, and the result is shown as a badge. Nothing is sent anywhere else. - Choose between three models: classical (fast, default), bert (DistilBERT) and guard (ModernBERT), if your server has them loaded. - Turn checking on or off at any time from the toolbar popup. Works on chatgpt.com, chat.openai.com, claude.ai and gemini.google.com. Privacy: your API key is stored only on your device. The extension contains no analytics, no ads and no remote code. Limitations: detection is statistical and can be wrong in both directions; treat the badge as a hint, not a guarantee. English only. PromptGuard is an independent open-source project and is not affiliated with, endorsed by or sponsored by OpenAI, Anthropic or Google. Source code: https://github.com/Arshu-1104/AI-Prompt-Injection