#12 Ethics Washing in AI: What Is Ethics Washing? And Why Does It Matter?

As many developers of artificial intelligence have been promising, AI is indeed proving to be highly disruptive. Almost daily, there are headlines about AI increasing the pace of research and innovation, which is also leading to harm, including job loss, discrimination, public surveillance, harassment, psychosis, suicide, and more. These disruptions are generating legitimate worries about whether AI is being developed and deployed in a responsible and trustworthy manner, and this has led to many attempts at regulation.

AI companies recognize these concerns, and they have responded in different ways. Some have made good faith attempts to voluntarily implement guardrails and protections that identify and mitigate risks. In other cases, however, AI organizations have attempted to assuage fears by advertising commitments to ethics, responsibility, and trustworthiness — and then failed to follow through on those commitments. These are cases of ethics washing.

Ethics washing is one of many types of washing, including whitewashing, greenwashing, pinkwashing, AI washing, and others. While the term whitewashing goes back several hundred years, the others are more recent, and they all attempt to highlight and expose misleading speech, behavior, and practices, which create the appearance that things are cleaner, more virtuous, more advanced, or normatively better than they are. While it would be interesting to conceptually analyze these various washings to characterize what they all share, the focus of this post is on ethics washing in AI and why it should concern us.

Some of the first discussions of ethics washing in AI occurred in 2019. That year, the European Commission’s 52-member High-Level Expert Group on Artificial Intelligence (HLEG-AI) published the Ethics Guidelines for Trustworthy AI, which has been highly influential in European and international discussions of AI ethics, governance, and regulation.

Thomas Metzinger, a philosopher at the Johannes Gutenberg University of Mainz in Germany and member of HLEG-AI, wrote an editorial for the German newspaper Der Tagesspiegel entitled “EU Guidelines: Ethics Washing Made in Europe.” Metzinger alleged that, even though the HLEG-AI was charged with producing ethics guidelines, it did not take ethical concerns seriously enough. The group had too few ethicists and too many industry representatives. The published guidelines ended up downplaying concerns about ethical harms; red lines that were included in early drafts were removed from the final document, which was “lukewarm, shortsighted, and deliberately vague.”

According to Metzinger, all of this — making a show about ethical deliberation, while producing results that are watered down and vague — was an intentional strategy on the part of industry. While he does not provide a precise definition of ethics washing, the following statement of his comes close: “Industry organizes and cultivates ethical debates to buy time — to distract the public and to prevent or at least delay effective regulation and policy-making.”

Near the end of 2019, Karen Hao wrote an article for MIT Technology Review entitled, “In 2020, Let’s Stop AI Ethics-Washing and Actually Do Something.” The main occurrence that led to her article was a now-infamous fiasco at Google involving the creation and almost immediate dissolution of an ethics board, the Advanced Technology External Advisory Council (ATEAC). This council was advertised as a way of advancing Google’s AI Principles and ethical deliberation about complex technologies, but it was dissolved after only nine days, in part due to the controversies about board membership.

According to Hao, this episode was “the most acute example” of ethics washing in AI. While she also does not provide a precise definition, her discussion suggests a slightly different one than Metzinger’s. She argues that organizations are engaging in a lot of “talk” about ethics, but “talk is just that — it’s not enough. Few companies can show tangible changes to the way AI products and services are evaluated and approved.” In her account, ethics washing amounts to talking about ethics that is mere lip service, and does not amount to real change. For her, the ATEAC episode raised concerns about ethics washing because it suggested that while companies might be quick to advertise commitments to ethics and responsibility, they are not necessarily invested in following through in a way that leads to real change.

What, precisely, is ethics washing?

In a recent paper, “How Can We Know if You are Serious? Ethics Washing, Symbolic Ethics Offices, and the Responsible Design of AI Systems,” my colleagues and I reviewed the academic literature on ethics washing and proposed the following definition: Signaling a commitment to ethics, without acting in ways that are sufficiently aligned with that signaling.

On this account, determining whether an individual or organization is engaging in ethics washing requires, at least, three tasks, including:

Identifying what it is that they are signaling or advertising — their talk about ethics.
Identifying what they are doing, such as what actions they are taking to facilitate responsible AI (e.g., organizational changes, policy developments, educational initiatives, etc.).
Determining the extent to which their actions align with their talk.

This definition of ethics washing does not require the identification of organizational intentions. An organization that is engaging in ethics washing might advertise commitments to ethics in an effort to “distract the public and to prevent or at least delay effective regulation and policymaking,” as Metzinger suggests. It might be trying to gain customers, or increase market share, or satisfy the demands of funding agencies. In some cases, the intentions might not be clear or might conflict.

AI organizations are complex

They consist of different offices or units with different aims, and in many cases, these aims might conflict. Some units might prioritize technological innovation. Others might focus on revenue generation. Still others might emphasize ethics and responsible development. It is not uncommon for internal conflicts to arise between different units over which aims should take precedence. In these cases, it can be difficult, if not impossible, to determine the overall ethical intentions of the organization.

I suspect that many AI organizations that signal commitments to ethics and responsibility have well-meaning, well-intentioned individuals who genuinely care about these issues. These organizations might even have entire ethics offices staffed with well-intentioned individuals. But what ultimately matters is organizational behavior — what the organization does with respect to the development and deployment of AI systems. If an organization talks about ethics and its actions fall far short of its talk, then it is engaging in ethics washing. Actions, not stated intentions, matter most.

So why does this matter?

Why is it valuable to define ethics washing and how to identify it? While there are many reasons, I will focus on one, relating to trustworthiness and accountability.

Given the importance of AI for the future of our world, it is vital that the organizations developing and deploying AI are worthy of public trust. AI is transforming our lives, and the organizations that create and profit from these systems should be trustworthy. But if they are to be trustworthy, they must also be accountable. Accountable to what or whom?

In the U.S., at least at the national level, there has been little appetite for enacting AI regulations. Some states, notably California and Colorado, have passed regulations that attempt to create guardrails that protect the public; however, there has been no significant AI regulation passed at the federal level. (See AI Blog by Karen Lindsley, Federal AI Policy Is Evolving Under the Trump Administration. What about Guardrails?)

As it relates to guardrails that facilitate responsible AI, there is no federal regulation in the U.S. to which AI companies are accountable. The situation is different in the European Union. (See AI Blog by Anne-Elisabeth Courrier: Strengths and Vulnerabilities of the European Union’s AI Act: Part I.) The U.S. way, moreover, is how many AI organizations want it. Many (though not all) argue that they can be trusted to voluntarily institute guardrails necessary to protect the public. Those companies that argue for self-regulation assert that they do not need to be held accountable to federal or state regulations. They claim that they will develop and deploy AI responsibly and that they will be accountable to their word.

Closing

It is important to identify instances of ethics washing, because organizations that engage in it are giving us reason to think that they are, in fact, not accountable to their word — and hence not worthy of trust. If AI companies argue that they do not need governmental regulation and are capable of regulating themselves, then they should be held to that standard. Talking a "big talk" about ethics and then failing to follow through is an indication that they are not capable of regulating themselves — at least not enough to be worthy of public trust.

Author

— by Justin B. Biddle, PhD, associate professor in the School of Public Policy at the Georgia Institute of Technology and a member of the Regulatory Knowledge and Support Program of the Georgia CTSA, 7/2026

Comments