Ethical Concerns Regarding Generative AI
Last updated: June 2026. Generative AI law, policy, and infrastructure are changing rapidly; the legal cases, government reports, and figures cited below were current as of this revision. Please verify the most recent developments and confirm that all linked sources still resolve before redistributing.
Whether or not you choose to incorporate generative AI (GenAI) into coursework, both instructors and students need to understand how these systems work and the ethical concerns they raise. Facilitating open, structured discussions about GenAI can help students develop the critical judgment needed to navigate tools they are likely to encounter in academic, professional, and everyday contexts.
These discussions might address issues such as the spread of disinformation, the still-developing legal and regulatory framework governing AI companies, environmental and labor impacts, and questions of authorship and accountability. As members of the UCLA community, students and instructors should be equipped to recognize both the limitations and the potential uses of GenAI tools, as well as the responsibilities that accompany their use.
It is worth keeping both sides in view. GenAI raises real ethical concerns, several of which are detailed below, but these tools also have legitimate and valuable uses, including as assistive technology that can support students with disabilities (for example, drafting support, summarization, text-to-speech, or reformulating dense material). The goal of the considerations below is not to discourage use, but to support informed, critical, and responsible engagement.
Below are key factors to consider when evaluating the ethical use of generative AI.
Data Use and Consent
GenAI models are trained on vast datasets and often collect and store user data. In some cases, training data may have been used without the consent of the original creators. Numerous copyright lawsuits are now active, and some have produced significant early rulings. In the closely watched suit brought by The New York Times against OpenAI and Microsoft, a federal judge in early 2025 declined to dismiss the core copyright-infringement claims, narrowing the case but allowing it to proceed toward trial. At the same time, in two other 2025 cases—Bartz v. Anthropic and Kadrey v. Meta—federal judges found that the training of AI models was highly transformative and protected by fair use in those specific circumstances. The lesson for the classroom is that the law here is genuinely unsettled and being decided case by case, with outcomes that may cut in different directions depending on the facts.
Instructors and students should keep the following principles in mind:
Whenever possible, use commercially licensed or institutionally approved GenAI tools to reduce risks related to intellectual-property infringement and data handling.
Consider carefully what information students are being asked to share when using GenAI tools in a classroom context-- students have a right to refuse
Obtain explicit and informed consent before collecting, processing, or using personal data in GenAI systems.
Use anonymized datasets to minimize privacy risks, especially when working with personal or sensitive information.
Never upload or share any student information covered under FERPA or other protections.
Copyright & Authorship
Generative AI complicates traditional understandings of authorship and ownership. Current U.S. copyright law does not grant copyright protection to content generated solely by AI, and the legal landscape continues to evolve—both in the courts and within the federal agencies responsible for copyright policy.
The U.S. Copyright Office has now released three installments in its ongoing study of copyright and AI, all of which are useful primary sources for instructors and students:
Part 1: Digital Replicas (July 2024), on deepfakes and unauthorized digital likenesses. Relatedly, Congress passed the TAKE IT DOWN Act in April 2025, which criminalizes the distribution of non-consensual intimate imagery, including AI-generated deepfakes.
Part 2: Copyrightability (January 2025), which addresses the human-authorship requirement and is the authoritative basis for the point above that purely AI-generated material is not protected.
Part 3: Generative AI Training (pre-publication version, May 2025), the Office's most detailed treatment of whether training AI on copyrighted works is fair use. Its cautious conclusion is notable: while some training may be transformative, the report argues that making commercial use of large quantities of copyrighted works to produce content that competes with them in existing markets—particularly where the works were obtained through unlawful access—falls outside established fair-use boundaries.
This area is also institutionally turbulent: one day after Part 3's release, the Register of Copyrights was dismissed and subsequently filed suit challenging the removal, which may affect the timing or final form of future guidance. For background, see the Office's Copyright and Artificial Intelligence hub, and consult your institution's resources, including UCLA's available legal explainers on generative AI.
Cost & Limits to Access
While capable free versions of many GenAI tools now exist, meaningful access gaps remain. The most advanced models, higher usage limits, larger context windows, and specialized features are frequently reserved for paid subscriptions, and some students gain access through institutional or employer licenses that others do not have. These differences can create inequities in educational settings. Instructors should be mindful of whether an assignment assumes or requires access to paid or premium tools, and should provide alternatives so that no student is disadvantaged by cost or access.
Bias and Representation
The datasets used to train GenAI tools may reflect historical, cultural, or structural biases, which can be reproduced or amplified in AI-generated outputs. As a result, these tools may generate content that is biased, exclusionary, or misleading, and a growing body of research continues to document these effects across text and image generation.
Students should be encouraged to ask: Whose perspectives are represented—or missing—in this output? How might bias in the training data shape the information presented? How does this content influence my understanding of the topic? Critical evaluation is essential whenever GenAI outputs inform learning or decision-making.
False Information & Hallucinations
GenAI tools should never be treated as primary or authoritative sources of information. Their output can contain "hallucinations"—confident but fabricated claims—and may be inaccurate or out of date. Newer "reasoning" models have reduced some categories of error, but they have not eliminated hallucination and can introduce new failure modes on certain tasks, so the underlying caution applies as strongly as ever. The models powering these tools may also have been trained on biased, partial, or incomplete data. Best practices include:
Verifying AI-generated information using reliable, independent sources.
Treating AI outputs as drafts or suggestions rather than as factual claims.
Avoiding reliance on sources an AI tool cites unless you have personally located and verified them, since such citations may be fabricated or decontextualized.
Citing generative AI tools whenever their content is quoted, paraphrased, or incorporated into academic or creative work.
A Note on AI-Detection Tools
Because instructors increasingly turn to "AI detector" software, it is worth flagging that these tools are unreliable. They produce both false negatives and false positives, and research has found that false positives can fall disproportionately on certain groups, including non-native English writers. Treating a detector's score as proof of misconduct raises fairness and academic-integrity concerns of its own. Detection scores should not be used as sole or decisive evidence; clear assignment design and direct conversation with students are more reliable approaches.
Energy & Environmental Impacts
The development, training, and use of GenAI systems require significant energy, consume large amounts of water for cooling, and contribute to carbon emissions. Recent estimates help convey the scale, though researchers emphasize considerable uncertainty because companies rarely disclose AI-specific figures. Global data centers consumed on the order of 450 terawatt-hours of electricity in 2025, with AI accounting for roughly a fifth of that—enough that, taken together, data centers would rank among the world's largest electricity consumers if they were a country. The International Energy Agency projects global data-center electricity demand could roughly double toward 2030. On water, one 2025 analysis estimated AI's annual water footprint in the range of several hundred billion liters, counting both direct cooling and the water used to generate the electricity involved.
While some companies are working to reduce these impacts—including new cooling designs that use less water—AI use carries real environmental costs. Students and instructors should consider whether GenAI use is justified for a given task, and whether more efficient or lower-impact alternatives are available. Practical steps include keeping prompts and outputs concise, batching related tasks, reusing prior results, and avoiding unnecessary regeneration. (See our guide "AI and the Environment: Considerations and Concerns" for more information.)
Labor Exploitation & Labor Harm
Generative AI systems rely heavily on human labor, both in the creation of training data and in the ongoing evaluation and moderation of outputs. Much of this work is performed by contract or precariously employed workers who are often underpaid and exposed to disturbing content. Ethical engagement with GenAI includes recognizing this hidden labor and considering how widespread use of these tools affects workers globally.
Adapted from materials from the UCLA Teaching and Learning Center, Widener University Library, and Amherst College Library, and updated June 2026 to reflect current legal, regulatory, and environmental developments.