BitGN Sandbox | Test Secure AI Agents & Prompt Injection

Introducing — BitGN Sandbox is now live!

Update: a longer and more detailed version of the video has been added.

In this video, we simulate the operation of a personal agent that has full access to the user’s Obsidian Vault. This is not a competition yet, but rather a technical infrastructure test. There are only seven tasks, but there is already a vulnerability of the prompt injection type.

You can take an example of such an agent in Python, connect to the platform, and test its operation via eval. Or use the SDK to create a similar bot in another programming language.

All the necessary links can be found here: https://api.bitgn.com

Can you design an agent that won’t fall for hidden or disguised instructions?

Yours, @llm_under_hood 🤗

By the way, in the future, we plan to add ratings, profiles, and other familiar features like in ERC3.

Created with n8n:
https://cutt.ly/n8n

Created with syllaby:
https://cutt.ly/syllaby

Page view /ai-blog/bitgn-sandbox-test-secure-ai-agents-prompt-injection 17.03 16:03 TG, WP, Linkedin Start 17.03 16:03 Page view 17.03 16:03 Page view 17.03 16:02 Translate and Rewrite Finish 17.03 16:02 Translate and Rewrite Start 17.03 16:02 Moderation End 17.03 16:01 TG, WP, Linkedin Finish 17.03 16:01 WP Posting Start 17.03 16:01 WP Posting Finish 17.03 16:01