Terminology.
US edition · Switch to the UK edition
- Acceptable use policy
- a short document that sets out which AI tools staff may use, for what, with which information, and what's not allowed.
- AI agent
- an AI system that can take actions on its own, such as sending emails, booking meetings or moving files, rather than only producing text.
- AI tools list
- a list of every AI tool your business uses, what it's for, what information it touches, who approved it and when it's next reviewed. Includes AI features built into software you already use. Sometimes called an AI register.
- Automated decision
- a decision about a person made by software, such as rejecting a job application. Several state laws set rules for these.
- Automation bias
- the tendency to trust an answer because a machine produced it, especially when it's well written.
- Bias audit
- an independent test of whether a tool's results differ unfairly between groups. Required each year for automated hiring tools used for New York City jobs.
- Business associate agreement (BAA)
- the contract HIPAA requires before a vendor handles protected health information for you.
- Business plan
- a paid AI plan for organisations, such as ChatGPT Business or Enterprise, Claude for Work or Microsoft 365 Copilot. Usually stops Microsoft training its AI on your content by default and comes with an agreement covering your data.
- Connector
- a link that gives an AI tool access to another system, such as your inbox, calendar, files or CRM.
- Consumer plan
- a free or personal paid AI account. Settings, retention and training use are controlled by the individual, not your business.
- Data minimisation
- using only the personal information you actually need for the task.
- Deepfake
- synthetic audio, images or video that convincingly imitate a real person.
- Disparate impact
- when a rule or tool that looks neutral disadvantages a protected group. Still unlawful under federal and many state laws.
- EEOC
- the Equal Employment Opportunity Commission, the federal agency that enforces workplace discrimination laws.
- FTC
- the Federal Trade Commission, which enforces the rules against deceptive and unfair business practices, including fake reviews and misleading AI claims.
- Generative AI
- AI that produces new content, such as text, images, audio or code, in response to a prompt.
- Grounding
- when an AI tool bases its answer on specific sources, such as your documents or a web search, rather than only on what it learned in training. Also called retrieval.
- Hallucination
- fluent, confident AI output that's false, such as an invented statistic, quote or citation.
- HIPAA
- the federal law that protects health information held by healthcare providers, health plans and the businesses that work for them.
- Human oversight
- a competent person reviewing AI output or decisions, with the time and authority to change or reject them.
- Large language model (LLM)
- the technology behind tools like ChatGPT, Claude, Gemini and Copilot. It generates text by predicting what's likely to come next.
- Model training
- using data to improve an AI model. Some consumer plans may use your conversations for this unless you switch it off.
- Personal information
- anything that identifies or relates to a person, such as a name, email address, résumé or interview notes.
- Prompt
- the instruction or question you give an AI tool.
- Prompt injection
- instructions hidden in content an AI reads, such as an email or web page, that manipulate it into doing something it shouldn't.
- Retention
- how long a provider keeps what you've entered. It varies by plan and settings.
- Risk assessment
- a written check of what could go wrong with a new use of personal information, and what you'll do about it. Required by several state privacy laws before high-risk processing.
- Sensitive personal information
- personal information that needs extra protection, such as government ID numbers, financial account details, precise location, health, racial or ethnic origin, religion, sexual orientation, citizenship or immigration status, genetic and biometric data, and children's data.
- Shadow AI
- AI tools used at work without the employer's knowledge or approval.
- State privacy laws
- the consumer privacy laws now passed in 24 states, starting with California's. They give people rights over their information and set rules for businesses above certain thresholds.
- Synthetic performer
- an AI-generated person in an ad. New York requires ads to disclose them.