“We start by teaching our AI right from wrong, filtering harmful content and responding with empathy.”
This analysis describes what OpenAI's agreement states, permits, or reserves. It does not constitute a legal determination about enforceability. Regulatory applicability and practical outcomes may vary by jurisdiction, enforcement context, and individual circumstances. Read our methodology
Characterizing filtering as part of teaching the AI 'right from wrong' establishes that harmful content controls are embedded in the model's foundational training, not only applied at the surface level.
Interpretive note: The excerpt is informal and figurative ('teaching our AI right from wrong'). The phrase describes a process but does not specify scope, methodology, or enforceability. The canonical claim captures the primary proposition; the empathy element is noted in omitted_material as a secondary but co-equal stated objective.
Users interact with AI systems that have been trained to filter harmful content and respond with empathy before reaching them.
How other platforms handle this
You agree that Your Content will not: be unlawful, harmful to adults or minors, threatening, abusive, harassing, tortious, defamatory, vulgar, obscene, profane, offensive, invasive of another's privacy, hateful...
The Services include safety features to block harmful content, such as content that violates our Prohibited Use Policy.
You agree that DoorDash may monitor and/or delete your User Content (but does not assume the obligation to do so) or may decide to not publish, display, or otherwise make available your User Content for any reason at DoorDash's sole discretion.
"We start by teaching our AI right from wrong, filtering harmful content and responding with empathy.Excerpt from OpenAI's Safety Overview
Get the research letter
Companies change their terms quietly. We read every version and catch what actually changed. One email a week on the changes that matter and what they mean.
Characterizing filtering as part of teaching the AI 'right from wrong' establishes that harmful content controls are embedded in the model's foundational training, not only applied at the surface level.
Users interact with AI systems that have been trained to filter harmful content and respond with empathy before reaching them.
ConductAtlas has identified this type of provision across 139 platforms. See the full comparison.
No. ConductAtlas is an independent monitoring service. We are not affiliated with, endorsed by, or sponsored by OpenAI.