TrojAI Browser Extension 作者: Andrew

The emergence of AI and specifically Large Language Models such as ChatGPT has taken the world by storm. The landscape is changing daily with more and more powerful LLMs released by the month.

While these technologies offer unprecedented increases in productivity for their users, they are not without risk. It is not clear if the inputs sent into the models are properly sanitized or used for future training. Researchers are constantly identifying new attack vectors that can circumvent safety measures baked into the LLMs and expose users to harmful content or induce data leaks.

TrojAI's LLM Monitor extension sits in between you and the LLM and monitors your inputs to ensure sensitive data is not exposed to the LLM. We've the designed the extension to integrate seamlessly into your current flow. You can interact with LLM's directly in the browser like you normally would and our extension will let you know if you are about to send sensitive information before the model is able to read it.

We are actively developing the extension and looking to rollout new features quickly. Our current capabilities include the following:
- Personally Identifiable Information (Credit Cards, SSN, emails, etc..)
- Toxicity (Words that could induce harmful content in a response from the LLM)
- Prompt Injection Detection
- Jailbreak detection
- DLP
- Multimodal protections and more.

Whether you're an individual looking to keep your own data safe, or an organization seeking to safely onboard LLMs into your developer workflow, TrojAI's cutting edge LLM monitoring extension is for you!