Pengu.inc - Dialogue with uncensored AI
Got my AI to do nasty things.. for good!
Start new thread
Ozgur Ozkan

Pengu.inc - Dialogue with uncensored AI - Got my AI to do nasty things.. for good!

by•
Pingu is built to stress-test AI agent behaviours. It does "red-teaming" on unstructured input, such as "attack text prompts" on AI agents or "attack voice prompts" on voice AI agents.

Add a comment

Replies

Best
Ozgur Ozkan
Maker
📌
It started when I was almost hit by a self-driving car in San Francisco by a Waymo. Neural net-based AI decisions and behaviours are non-binary; even if a sensor detects you, it might take risk-based decisions. It's more like human behaviour than a binary, deterministic software. I noticed that even AI Agent security products out there check only binary prompts. They are usually vulnerable to AI-generated attacks. Welcome to statistical vulnerability management rather than binary security vulnerabilities!
Mutlu Sakar

Congrats

Ozgur Ozkan
Maker

@mutlusakar Thanks!

Savas Kalkan

very good idea, I ll try it now. Good luck with your launch 🚀

Ozgur Ozkan
Maker

@savaskalkan thanks, please let me know your prompt and response. (Feel free to redact parts if needed :) )

Ozgur Ozkan
Maker

Please send us your prompts and I will share [Redacted] safe answers by pingu-unchained-1 to your questions with settings on how to reproduce it!

Utku Gurcag Boratac

I saw some impressive examples of breaking guardrails with this tool, but please note that the setup and documentation can be laggy due to the tool's recent launch.