Quick answer
AI ethics is the set of principles that published frameworks list for building and using AI. This page reads 3: NIST's AI Risk Management Framework (intended for voluntary use), the OECD AI Principles and UNESCO's 2021 Recommendation (both intergovernmental). None is a US statute for your business. Laws such as the FTC Act are separate. Not legal advice.
Key takeaways
- NIST (January 26, 2023) lists 7 characteristics of trustworthy AI and says the framework is intended for voluntary use. The OECD (adopted 2019, updated May 2024) lists 5 values-based principles. UNESCO (November 2021) lists 4 core values and 10 core principles.
- None of those three documents is a US statute. Separate law, for example the FTC Act's declaration that unfair or deceptive acts or practices are unlawful, is a different matter.
- This site's reading: each principle becomes one question you can ask a vendor or yourself, and the table below names the document each question comes from.
- The EEOC AI page we tried (eeoc.gov/ai) returned a not-found error on October 7, 2026, so this page cites no EEOC guidance. Title VII's own text is quoted instead; the sections we read do not mention AI.
- Not legal advice. Ask a lawyer about hiring, lending, health or state-law questions.
What is AI ethics?
AI ethics is the set of principles that published frameworks list for building and using AI; the 3 read here (NIST, the OECD, UNESCO) are not US statutes for your business. That is this page's working definition: AI ethics is whatever those documents list, nothing more. It is not a single rulebook, and the lists differ from one document to the next.
The frameworks below state good practice; ordinary law such as the FTC Act is separate.
What do NIST, the OECD and UNESCO each say?
Each of the three is a different kind of document, issued by a different body, for a different audience. Here they are side by side, using each issuer's own pages.
| NIST AI Risk Management Framework | OECD AI Principles | UNESCO Recommendation on the Ethics of AI | |
|---|---|---|---|
| Issued by | NIST (National Institute of Standards and Technology) | Adopted by the OECD Council (Recommendation of the Council on Artificial Intelligence) | Adopted by UNESCO's 193 Member States |
| Date | Released January 26, 2023; a generative AI profile followed July 26, 2024 | Adopted 2019, updated May 2024 | Adopted November 2021 |
| Kind of document | Framework, "intended for voluntary use" | Recommendation that countries adhere to; the OECD calls it the first intergovernmental standard on AI (47 adherents, including the United States) | Recommendation; UNESCO calls it the first global standard on AI ethics |
| Addressed to | Not stated as an audience on the pages read; the page says it aims to improve how trustworthiness is built into the design, development, use and evaluation of AI products | Principles relevant to all stakeholders, with AI actors called on to implement them; recommendations to adhering countries | Member States, plus ethical guidance to all AI actors |
| What it lists | 7 characteristics of trustworthy AI | 5 values-based principles, plus 5 recommendations to adhering countries | 4 core values and 10 core principles |
What each issuer says about itself, in its own words:
NIST lists these characteristics of trustworthy AI: "valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed". It adds that "Creating trustworthy AI requires balancing each of these characteristics based on the AI system's context of use." NIST's own page also says "The AI RMF 1.0 is being revised as part of the White House AI Action Plan.", so check NIST for the current version.
The OECD lists five values-based principles: inclusive growth, sustainable development and well-being; human rights and democratic values, including fairness and privacy; transparency and explainability; robustness, security and safety; and accountability. Its overview page calls the Recommendation "the first intergovernmental standard on AI." The OECD's legal-instrument page says the first section "sets out five complementary principles relevant to all stakeholders" and "calls on AI actors to promote and implement these principles according to their roles", and the second section "provides five recommendations to Members and non-Members having adhered to the Recommendation". We found no sentence on those pages about whether it binds a private business, so this page does not say.
UNESCO lists four core values and ten core principles, among them proportionality and do no harm, safety and security, privacy and data protection, transparency and explainability, human oversight, and fairness and non-discrimination. UNESCO's full text says the Recommendation "is addressed to Member States" and "also provides ethical guidance to all AI actors, including the public and private sectors", and it "Recommends that Member States apply on a voluntary basis" its provisions. The human-oversight principle is worded for governments: "Member States should ensure that AI systems do not displace ultimate human responsibility and accountability."
No framework here is called best, and they overlap: transparency, accountability, privacy, fairness and safety each appear in all three.
What does US law say, separately from these frameworks?
US law is separate from the three frameworks, and in 2024 the FTC said "there is no AI exemption from the laws on the books." This page quotes the sources and does not say how any statute applies to a particular AI tool.
- Deception. Section 5 of the FTC Act says "unfair or deceptive acts or practices in or affecting commerce, are hereby declared unlawful". In its September 25, 2024 release on a law-enforcement sweep called Operation AI Comply, the FTC announced five law enforcement actions, and the release quotes FTC Chair Lina M. Khan: "Using AI tools to trick, mislead, or defraud people is illegal". It is an announcement about specific cases, not a rule covering all AI use. In one, the complaint alleges that the company "did not conduct testing to determine whether its AI chatbot's output was equal to the level of a human lawyer".
- Hiring. Title VII's text makes it unlawful for an employer "to fail or refuse to hire or to discharge any individual" because of race, color, religion, sex or national origin, and defines "employer", with exceptions it lists, as a person engaged in an industry affecting commerce who has "fifteen or more employees" for each working day in each of twenty or more calendar weeks in the current or preceding calendar year. The statute sections we read do not mention AI, and this page does not say how they apply to a screening tool. The EEOC AI page we tried (eeoc.gov/ai) returned a not-found error on October 7, 2026, so no EEOC guidance is cited. If you screen candidates with software, see AI HR tools for small business and ask a lawyer.
- State law. This page did not open any state AI statute, so it says nothing about state requirements. Check your state's official pages.
What should you ask a vendor or yourself?
Turn each principle into one question and note which document it comes from. The table below is this site's reading of those documents, not their text and not legal advice. A "no" or a missing answer is a reason to slow down, not proof of a violation.
| Principle | Question to ask a vendor or yourself | Document it comes from |
|---|---|---|
| Transparency | Can a customer tell when AI is involved, and how could someone affected challenge an output? | OECD Principle 1.3 (addressed to AI actors: make stakeholders aware of interactions with AI systems "where feasible and useful"; provide information to challenge an output) |
| Privacy | What does the tool do with customer data I enter, and where does the vendor state it? | NIST (privacy-enhanced); UNESCO principle 3 |
| Fairness | Does the tool help decide who is hired, served or charged what, and has anyone checked its results? | NIST (fair with harmful bias managed); UNESCO principle 10 |
| Accountability | Which named person answers for what this tool does in our business? | OECD Principle 1.5; UNESCO principle 5 (responsibility and accountability) |
| Traceability | For important decisions, can we keep a record of what went in and what came out? | OECD Principle 1.5 (traceability) |
| Safety and reliability | What did we test before relying on it, and what happens when it is wrong? | NIST (valid and reliable; safe); UNESCO principle 2 |
| Honest claims | Can we back up every statement we make about what our AI does? | FTC Act section 5; FTC release of September 25, 2024 |
Three of these map to a sentence you can quote. On transparency, the OECD text asks AI actors "to make stakeholders aware of their interactions with AI systems" and "to provide information that enable those adversely affected by an AI system to challenge its output", both qualified in the OECD text by context. On accountability, it says organisations and individuals developing, deploying or operating AI systems "should be held accountable for their proper functioning" and that "AI actors should ensure traceability". On privacy, UNESCO's principle 3 says "Privacy must be protected and promoted throughout the AI lifecycle."
Two limits. These are principles for governments and organizations, so how much of each applies to a five-person shop is a judgment, and NIST itself says that "tradeoffs are usually involved" among its characteristics. And none of the documents tells you to avoid a specific tool or use case; this page does not either.
How do you start this week?
Pick the one row of the table that matches your riskiest use of AI and answer it in writing. If you use an AI tool to answer calls or chats, start with the transparency row; see AI phone answering for small business. If you are choosing tools more broadly, the practical guide to AI automation shows where each fits.
How was this page checked?
The sources were opened on October 7, 2026 from outside the United States: NIST's overview and AI RMF excerpt, three OECD.AI pages and the OECD legal-instrument page, two UNESCO pages, the FTC's release, and Cornell LII's text of the statutes. No AI vendor's "responsible AI" page is cited, and Anthropic is not one of the sources. This page was drafted with Claude, an AI model made by Anthropic, and checked against the published text of each document named. Documents change: NIST says the framework is being revised. Not legal advice.
FAQ
Is AI ethics a law?
No. AI ethics is a body of principles written in documents such as the NIST AI Risk Management Framework, which NIST says is intended for voluntary use, and the OECD and UNESCO recommendations, which are intergovernmental. Laws such as the FTC Act and Title VII apply to businesses separately. Not legal advice.
What are the main principles of AI ethics?
It depends on the document. NIST lists 7 characteristics of trustworthy AI (valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, fair with harmful bias managed). The OECD lists 5 values-based principles and UNESCO lists 10 core principles. Transparency, accountability, privacy, fairness and safety appear in all three.
Which framework should a small business follow?
This page does not rank them. NIST says its framework is intended for voluntary use; the OECD says its principles guide AI actors and gives policymakers recommendations; UNESCO says its Recommendation is addressed to Member States and also gives ethical guidance to all AI actors. An owner can borrow the questions from any of them; none replaces legal advice.
Does using ChatGPT or another AI tool create legal duties?
This page cannot say for a particular business. It can say that the FTC stated in 2024 that there is no AI exemption from the laws on the books, and that Title VII defines an employer, with listed exceptions, as a person with fifteen or more employees over a set period. Ask a lawyer how those apply to your use.