Skip to content
Last checked: October 2026. AI laws are changing quickly, so check official sources before relying on any date.
AI ethics is the set of principles that guide how AI is built and used so it is fair, safe, transparent and accountable. It matters more every year, as AI moves from answering questions to taking actions for people. This guide covers the core principles, real examples of what goes wrong, the main laws and frameworks, and practical steps any team can take.
The core principles of AI ethics
| Principle | What it means in practice |
|---|---|
| Fairness | The system should not treat people worse because of race, gender, age or other protected traits |
| Transparency | People should know when they are dealing with AI and how decisions are made |
| Accountability | A named person or team is responsible for what the AI does |
| Privacy | Personal data is collected only with good reason and protected |
| Safety and reliability | The system works as intended and fails safely |
| Human oversight | People can review, correct or override important AI decisions |
Most major frameworks use some version of this list, including the OECD AI Principles (2019, updated 2024) and UNESCO’s Recommendation on the Ethics of AI, adopted by all 193 member states in 2021.
Real examples of AI ethics failures
Amazon’s recruiting tool
Amazon began building an AI tool to rank job applicants in 2014. By 2015 it learned the tool was penalizing resumes that mentioned “women’s,” as in “women’s chess club,” because it was trained on ten years of mostly male resumes. The team was disbanded in early 2017, and Reuters reported the story in 2018.
COMPAS risk scores
In 2016 ProPublica analyzed COMPAS, software used by US courts to predict reoffending. It found Black defendants who did not reoffend were almost twice as likely as white defendants to be wrongly labeled high risk. The vendor disputed the findings, and the case is still a standard example of bias in automated decisions.
The Apple Card investigation
In 2019, viral posts claimed the Apple Card gave women lower credit limits than their husbands. New York’s financial regulator investigated and, in 2021, found no fair-lending violations, but criticized the companies for poor customer service and a lack of transparency. The lesson: unexplained decisions damage trust even when they are not illegal.
IBM Watson for Oncology
Internal IBM documents reported in 2018 showed Watson for Oncology made unsafe treatment suggestions, partly because it was trained on hypothetical cases rather than real patient data. IBM sold its Watson Health assets in 2022.
Clearview AI
Clearview AI scraped billions of photos from the web to build a facial recognition database. European regulators have fined it repeatedly, including €30.5 million from the Dutch data protection authority in 2024.
Recent cases
- Grok deepfakes (2025–2026): users of X’s Grok chatbot generated sexualized images of real people without consent. Several countries blocked it, and the European Commission opened an investigation in January 2026.
- AI companion chatbots and teens: lawsuits over teen self-harm led Character.AI to block open-ended chats for under-18 users, and the companies involved reached settlements in principle in January 2026. The FTC opened an inquiry into AI companion chatbots in 2025.
AI laws and frameworks today
EU AI Act
The world’s first broad AI law entered into force in August 2024 and applies in stages:
| Date | What applies |
|---|---|
| February 2, 2025 | Bans on unacceptable-risk AI, plus an AI literacy duty |
| August 2, 2025 | Rules for general-purpose AI models |
| August 2, 2026 | Transparency duties, such as disclosing chatbots and labeling deepfakes |
| December 2, 2026 | New bans on AI “nudifier” apps and AI-generated sexual abuse material |
| December 2, 2027 | Rules for high-risk AI such as hiring and credit scoring (delayed from 2026) |
| August 2, 2028 | Rules for high-risk AI built into regulated products |
The delays came from the EU’s “Digital Omnibus” package, which took effect in July 2026.
United States
There is no single federal AI law. The federal government has focused on removing barriers to AI and, since December 2025, on challenging state AI laws it sees as burdensome. Congress has not passed a law overriding state rules, so state laws still apply. Notable examples include California’s SB 53, which requires large AI developers to publish safety frameworks, and California’s SB 243 on companion chatbots, both in effect since January 2026. Colorado replaced its original AI Act with a narrower law that takes effect on January 1, 2027.
Voluntary frameworks
- NIST AI Risk Management Framework (2023): a US guide built around four steps: govern, map, measure and manage.
- ISO/IEC 42001 (2023): the first certifiable standard for an AI management system.
- Company principles: Microsoft’s six principles are fairness, reliability and safety, privacy and security, inclusiveness, transparency and accountability. Google rewrote its AI principles in 2025 and removed its earlier list of uses it would not pursue.
How to put AI ethics into practice
- Know where you use AI. Keep a simple list of every AI tool and what decisions it touches.
- Rate the risk. Hiring, lending, health and anything affecting children needs the most care.
- Test for bias. Check results across groups before launch and on a regular schedule.
- Tell people. Disclose when customers are talking to AI or reading AI-generated content.
- Keep a human in the loop. Let people review and appeal important decisions.
- Protect data. Share only what the tool needs, and check where vendors store it.
- Assign an owner. Make one person or team responsible for each AI system.
- Watch AI agents closely. Agents that take actions, such as sending email or spending money, need limits, approvals and logs.
FAQs
What are the main principles of AI ethics?
Fairness, transparency, accountability, privacy, safety and human oversight.
Is AI ethics required by law?
Partly. The EU AI Act makes many of these principles legal duties, and some US states have their own AI laws. Elsewhere, most AI ethics is still voluntary.
Why does AI become biased?
Usually because it learns from historical data that reflects past unfairness, as in Amazon’s recruiting tool, or because it is tested on too narrow a group of people.
Choosing AI agents for your business? See how Klyvoo reviews AI agents, including privacy and data handling.