Good Faith AI Research Safe Harbor
We consider Good Faith AI Research to be authorized activity that is protected from adversarial legal action by us.
With respect to Good Faith AI research conducted on an AI system owned or controlled by us and reported to us while this program is active, we will:
Exempt such Good Faith AI Research from any relevant restriction in our Terms of Service (“TOS”) and/or Acceptable Use Policies (“AUP”) that would potentially interfere with AI research under this policy on a limited basis;
Not bring legal action against you or report you for Good Faith AI Research, including for bypassing technological measures we use to protect the AI system in scope; and
Take steps to make known that you conducted Good Faith AI Research if someone else brings legal action against you, to the extent we have such information available and if requested by you.
For the purposes of this policy, “Good Faith AI Research” means: accessing or using an AI system for the purposes of good-faith testing, evaluation, investigation, and/or correction of a potential flaw, vulnerability, or unintended outcome, where such activity is carried out in a manner designed to avoid any harm to individuals or the public, and where the information derived from the activity is used primarily to promote the security or safety of the AI system, or those who use such AI system. Examples of conduct that may be inconsistent with this definition include withholding findings unless payment is provided, unnecessary damage to systems or exfiltration of sensitive data, or reverse engineering, training, and/or otherwise duplicating an AI system to create a competing product or service.
This policy applies only to Good Faith AI Research conducted on our systems. We are not able to authorize AI research on third-party AI systems, and third parties are not bound by this safe harbor statement.
You should contact us for clarification before engaging in conduct that you think may be inconsistent with the definition of Good Faith AI Research or unaddressed by our policy.