zihao-ai/unthinking_vulnerability

To Think or Not to Think: Exploring the Unthinking Vulnerability in Large Reasoning Models

17
/ 100
Experimental

This project identifies and explores a critical 'Unthinking Vulnerability' in Large Reasoning Models (LRMs), where specific inputs can bypass their intended reasoning processes. It provides tools to either maliciously exploit this flaw to induce incorrect outputs or beneficially monitor and enhance the safety and efficiency of these models. AI developers and researchers working with LRMs would use this to build more robust and secure AI systems.

No commits in the last 6 months.

Use this if you are an AI developer or researcher concerned about the reliability and security of Large Reasoning Models and need tools to test for and mitigate reasoning bypass vulnerabilities.

Not ideal if you are looking for an end-user application or a pre-built solution for a specific business problem, as this is a research toolkit for model developers.

AI-safety Large-Language-Models AI-vulnerability-testing Machine-Learning-security Model-auditing
No License Stale 6m No Package No Dependents
Maintenance 2 / 25
Adoption 7 / 25
Maturity 8 / 25
Community 0 / 25

How are scores calculated?

Stars

33

Forks

Language

Python

License

Last pushed

May 21, 2025

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/agents/zihao-ai/unthinking_vulnerability"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.