dropbox/llm-security
Dropbox LLM Security research code and results
This project helps security researchers and AI safety engineers understand and demonstrate 'prompt injection' attacks against large language models like ChatGPT. By crafting specific inputs with repeated tokens, you can observe how an LLM's intended behavior can be overridden, potentially leading to unintended responses or even data leakage. It's designed for those who work on securing AI applications and need to validate system robustness.
256 stars. No commits in the last 6 months.
Use this if you are an AI security researcher or engineer responsible for identifying and mitigating vulnerabilities in LLM-powered applications.
Not ideal if you are looking for a general-purpose LLM development tool or a solution for common application-level prompt engineering.
Stars
256
Forks
29
Language
Python
License
Apache-2.0
Category
Last pushed
May 21, 2024
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/llm-tools/dropbox/llm-security"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
ethz-spylab/agentdojo
A Dynamic Environment to Evaluate Attacks and Defenses for LLM Agents.
guardrails-ai/guardrails
Adding guardrails to large language models.
JasonLovesDoggo/caddy-defender
Caddy module to block or manipulate requests originating from AIs or cloud services trying to...
inkdust2021/VibeGuard
Uses just 1% memory while protecting 99% of your personal privacy.
deadbits/vigil-llm
⚡ Vigil ⚡ Detect prompt injections, jailbreaks, and other potentially risky Large Language...