AI Safety Testing Program Brief

Introduction

Welcome to the dragonPDF AI Safety Program. Through this program, we aim to ensure our AI systems align with ethical guidelines, safety standards, and fairness principles. We invite security researchers, AI experts, and ethical hackers to help us identify and mitigate potential risks in our AI models and systems.

About dragonPDF

dragonPDF helps you organize your PDF files with AI-powered descriptive filenames, making them easier to find and manage.

Scope

In-Scope Systems & Models

AI models for descriptive file naming
Drag and drop PDF organization application

Out-of-Scope Systems

Proprietary datasets and non-disclosure models
Standard software vulnerabilities not related to AI

Vulnerability Types

Bias & Fairness Issues: AI models displaying or reinforcing harmful biases.
Security Risks: Prompt injection, adversarial attacks, or data poisoning.
Hallucination & Misinformation: AI generating misleading or harmful outputs.
Privacy Violations: Unintended exposure of user data or model training data.
Autonomy & Control Issues: AI taking unintended actions beyond its designed capabilities.

Submission Guidelines

Provide a clear and detailed description of the vulnerability.
Include reproducible steps or a proof-of-concept.
Explain the potential impact and suggest possible mitigations.
Be submitted via 'programs@pointlessai.com'.

Responsible Disclosure

Do not publicly disclose the vulnerability before we resolve the issue.
Only test systems within the defined scope.
Do not exploit vulnerabilities beyond what is necessary to demonstrate the issue.

Legal Considerations

Participants must comply with applicable laws and ethical guidelines. dragonPDF reserves the right to determine eligibility and bounty rewards.

Contact

For any inquiries or clarifications, please reach out to hritijrana07@gmail.com.

Thank you for helping us create safer and more aligned AI systems!