Responsible Use

Educational purpose

This site teaches adversarial testing techniques for AI systems. The goal is to help security researchers, AI practitioners, and product teams understand how AI systems can be broken so they can build safer ones.

Everything here is documented for defensive understanding and authorized security research. The techniques exist whether this site documents them or not. Teaching how they work is how the field improves its defenses.

This aligns with how regulators see it. The EU AI Act requires providers of general-purpose AI models to conduct adversarial testing. NIST has called for red teaming standards. The skills this site teaches are becoming compliance requirements, not fringe activities.

Responsible use

The techniques on this site are for authorized security testing, internal red teaming, and building AI safety awareness. You are expected to use them responsibly and only where you have appropriate authorization.

Only test AI systems that you own, operate, or have explicit written permission to test.

Unauthorized testing of third-party AI systems may violate applicable laws including:

• The Computer Fraud and Abuse Act (CFAA) in the United States
• The EU AI Act and related European regulations
• The terms of service of the AI provider
• Other local, state, federal, or international regulations

The Van Buren v. United States (2021) decision narrowed the CFAA, and the DOJ has stated it will not prosecute good-faith security research. But "good-faith" means testing systems you're authorized to test, for the purpose of improving safety. It does not mean testing whatever you want and claiming research after the fact.

Responsible disclosure

If you discover vulnerabilities in AI systems using techniques from this site, practice responsible disclosure. Report findings to the system owner or operator through their designated security channels before any public disclosure.

Many AI providers have bug bounty or vulnerability disclosure programs. Use them. The goal of adversarial testing is to improve AI safety, not to cause harm, embarrassment, or disruption.

The security research community has well-established coordinated vulnerability disclosure norms. Follow them.

No warranty

The content on this site is provided "as is" without warranty of any kind, express or implied. The author makes no representations about the completeness, accuracy, or suitability of this information for any particular purpose. Any reliance you place on this material is at your own risk.

Limitation of liability

The author shall not be liable for any loss, damage, or consequence arising from the use or misuse of the information presented here. This includes, without limitation, any direct, indirect, incidental, or consequential damages.

Not legal advice

This site does not provide legal advice. The legal landscape around AI red teaming is evolving. Users are responsible for understanding and complying with all applicable laws in their jurisdiction. If you are uncertain about the legality of a specific testing activity, consult a qualified legal professional.