Acceptable Use Policy
Prohibited scenarios, requirements for working with open data and restrictions for high-risk areas.
1. General principle
Use Nola legally, fairly, and with respect for security, copyrights, privacy, and the terms of third-party services. You cannot use agent autonomy to scale harm.
2. Illegal and dangerous content
- Promoting crime, violence, terrorism, human trafficking, weapons or dangerous substances.
- Sexual exploitation of minors and any sexualized interactions with children.
- Encouraging self-harm or instructions that pose an immediate threat to life.
- Fraud, extortion, phishing, document forgery, or circumvention of sanctions and mandatory restrictions.
3. Cybersecurity
- Hacking, malicious code injection, credential theft, accessibility violations, and unauthorized scanning are prohibited.
- Protective research and testing is only permissible on proprietary systems or with the express permission of the owner and agreed upon boundaries.
- You cannot bypass Nola's security mechanisms, hide a malicious target, or use prompt injection to obtain someone else's data.
4. Privacy and open source search
- Public availability of data in itself does not negate the requirements of the law, the rights of the subject and the restrictions of the source.
- Harassment, doxxing, creating bases for discrimination, searching for sensitive data without a legitimate purpose, and de-anonymizing individuals are prohibited.
- You cannot access closed accounts, paid databases without a license, leaks or data protected by authorization.
5. Deception and manipulation
- AI should not be impersonated by a real person to mislead or hide automation where disclosure is required.
- Unauthorized impersonation of identity, fraudulent reviews, mass spam and manipulation of socially significant processes are prohibited.
- Synthetic materials must be labeled when failure to label them would be likely to mislead a reasonable person.
6. High-risk areas
Nola can help collect and organize information, but should not independently make final decisions in medicine, law, finance, insurance, employment, education, housing and law enforcement. Requires skilled personnel and significant human supervision.
7. Response measures
We can stop the task, limit tools, request additional confirmation, save necessary security logs, or block access. The assessment takes into account context, intent, scope, repeatability, and remediation.