usage policy
In plain wordsA company's published rules on what people may and may not use its AI for — for example, no help building weapons, no scams and no sexual content involving children.
A provider's published rules governing what customers and end users may do with its AI models and services. Typical policies prohibit using the service for weapons development (above all weapons capable of mass casualtiesCBRN risk, n.: The danger that AI could help someone make or use chemical, biological, radiological or nuclear weapons — for example, by explaining hard technical steps an attacker could not work out alone.), malware, child sexual abuse material, fraud, harassment, election interference and some forms of surveillance; restrict higher-risk uses, such as medical, legal or financial advice or automated decisions about people, unless safeguards are in place; and impose duties on businesses building on the service, such as disclosing that users are talking to an AI. Policies are enforced through classifierssafety classifier, n.: A separate AI model that checks messages going into or coming out of a chatbot and flags or blocks those that break safety rules, like a security guard at the door., account-level monitoring, warnings, suspensions and contract terms — and, for open-weight modelsopen-weight model, n.: An AI model whose trained internal numbers are published for anyone to download and run on their own computers, even if the data and code used to make it stay private., only through license conditions. A usage policy constrains people; a model specificationmodel specification, n.: A public document in which an AI company spells out how its models are supposed to behave — what they should do, refuse and prioritize, and whom to obey when instructions conflict. describes how the model itself should behave, though the two are usually kept consistent.
The account was suspended for violating the usage policy on automated political messaging.
Origin
Adapted from the acceptable use policies of 1990s internet service providers. OpenAI's GPT-3 API launched in 2020 with stated prohibitions on harmful uses, and other commercial model APIs followed. Open-weight releases attached use restrictions through Responsible AI Licenses, as for BLOOM and Stable Diffusion in 2022, and through acceptable use policies such as the one attached to Meta's Llama 2 license (July 2023).
Usage note
Policies change often and differ between providers: a use prohibited by one may be allowed with safeguards by another. Not to be confused with a model licensemodel license, n.: The legal terms under which an AI model may be downloaded and used — for example, whether businesses may use it, whether changes may be shared, and which uses are banned., which may incorporate a usage policy but also governs copying and redistribution.
Related
- model specificationmodel specification, n.: A public document in which an AI company spells out how its models are supposed to behave — what they should do, refuse and prioritize, and whom to obey when instructions conflict.
- misusemisuse, n.: People deliberately using AI to cause harm — for scams, harassment, propaganda, hacking or weapons — as opposed to harm from AI making mistakes or pursuing goals of its own.
- safety classifiersafety classifier, n.: A separate AI model that checks messages going into or coming out of a chatbot and flags or blocks those that break safety rules, like a security guard at the door.
- structured accessstructured access, n.: Letting people use an AI system through a controlled channel, such as a website or programming interface, instead of handing over the model itself, so that the developer can limit misuse.
- model licensemodel license, n.: The legal terms under which an AI model may be downloaded and used — for example, whether businesses may use it, whether changes may be shared, and which uses are banned.
- refusalrefusal, n.: When an AI declines to do what it was asked — 'I can't help with that' — usually because the request seems harmful or breaks the rules it was trained to follow.
- model specificationmodel specification, n.: A public document in which an AI company spells out how its models are supposed to behave — what they should do, refuse and prioritize, and whom to obey when instructions conflict.