Generative AI Prohibited Use

This is Google's Generative AI Prohibited Use Policy, published on Google's Privacy & Terms site and last modified December 17, 2024. It is a usage policy, not a model card — it governs how users may interact with generative AI across Google products and services that reference it. No specific model name, version, or superseded predecessor is identified.

Not disclosed in this document. The policy is not a model card and contains no benchmark scores, modality descriptions, parameter counts, or context window information.

Not disclosed in this document. No information about how models are tested, evaluated, or assessed for compliance with these prohibitions is provided.

Not disclosed in this document. The policy enumerates prohibited output categories — including CSAM, violent extremism, non-consensual intimate imagery, self-harm facilitation, and malware — but describes no red-team exercises, catastrophic-risk evaluations, or threshold findings.

The policy prohibits users from "circumvention of abuse protections or safety filters" and from "manipulating the model to contravene our policies," indicating that such filters exist. No technical details about classifier thresholds, refusal training, ASL tiers, or access controls are described. Google reserves the right to make exceptions for educational, documentary, scientific, or artistic uses, or where "harms are outweighed by substantial benefits to the public."

The policy applies to all Google products and services that explicitly reference it. No API terms, licensing details, tiered access controls, or geographic restrictions are specified in this document.

The card does not discuss technical limitations of the underlying models. The only acknowledged boundary is the exception clause permitting otherwise prohibited uses when educational, documentary, scientific, or artistic merit applies, or when public benefit outweighs harm — but no process for invoking that exception is described.

The document carries a single modification date of December 17, 2024. No changelog, version history, or description of what changed relative to a prior version is included.