CONCEPT

Ethics Washing

Metzinger’s term for the performance of moral concern as a substitute for, and a shield against, actual moral constraint—the production of ethics guidelines engineered to look serious while binding no one, coined from his experience inside the European Commission’s AI ethics process.

Ethics washing is what happens when the institutions meant to constrain a technology are captured by the industry they were supposed to govern, and the result—the ethics board, the principles document, the voluntary commitment—becomes not a limit on harmful practice but a legitimating cover for it. Thomas Metzinger coined the term by analogy to greenwashing—the performance of environmental concern as a substitute for environmental action—and grounded it in direct observation. From 2018 to 2020 he served on the European Commission’s High-Level Expert Group on Artificial Intelligence, where he watched the language of red lines and non-negotiable principles systematically removed from official guidelines under industry pressure, and where the group charged with constraining AI was composed overwhelmingly of representatives from the industry it was meant to constrain. The final document contained no non-negotiable principles. Everything had become a matter of balance and trust. Mid-term and long-term risks—including the possible creation of artificially suffering minds—were quietly purged because acknowledging them would have disrupted the marketing narrative around AI. The ethics washing charge is not that the participants acted in bad faith; it is that the structure of the process guaranteed this outcome regardless of individual intentions. A group dominated by the regulated industry will produce guidelines that serve that industry, and calling the result “ethics” makes things worse rather than better by conferring legitimacy without imposing constraint.

In the [YOU] on AI Field Guide

The cycle returns repeatedly to the gap between what the institutions governing AI are supposed to do and what they actually do. Ethics washing is the name for one of the most important mechanisms producing that gap: the systematic conversion of genuine ethical constraint into its performative substitute. The companion volume argues that the people building these systems cannot outsource responsibility to oversight bodies; Metzinger’s analysis from inside one of those bodies confirms why: the bodies themselves are structurally designed to absorb and neutralize the responsibility rather than exercise it.

Ethics washing is particularly dangerous because its product—the ethics document, the principles statement, the governance framework—reassures the public, satisfies the press, and allows the companies and labs involved to claim moral seriousness while the underlying trajectory of development remains entirely unchanged. The appearance of ethics is in some ways worse than no ethics, because it forecloses the demand for the real thing. A toothless ethics process is not a failed attempt at ethics; it is a successful attempt at managing the demand for ethics.

The Alignment Problem as Central Challenge

Origin

The concept emerged directly from Metzinger’s experience on the EU HLEG-AI (High-Level Expert Group on Artificial Intelligence), which he described in a series of interviews and essays from 2019 onward, culminating in a detailed account in his 2021 paper on artificial suffering. The EU’s AI ethics guidelines—published in 2019 as “Ethics Guidelines for Trustworthy AI”—were, in his assessment, among the most substantial documents on the subject anywhere in the world, which made their compromises more revealing rather than less. He saw firsthand how a process with genuine experts and genuine intentions could produce an outcome that served industry interests, and his account identified the structural mechanisms responsible: the ratio of industry representatives to ethicists, the exclusion of long-term and existential risks from the mandate, and the political pressure to produce a document that would not frighten investment.

Metzinger has been explicit that his complaint is not against the individual members of the group, many of whom were well-intentioned, but against the institutional design that made ethics washing the predictable output. The design features that produce ethics washing are not accidental; they serve the interests of the parties that dominate the design process, which are the parties whose practices the ethics process was meant to constrain.

Key Ideas

Structural capture, not individual bad faith. Ethics washing does not require anyone to be dishonest. It requires only that the process be structured so that the parties with the most to lose from genuine constraint have the most influence over the process. The mechanism is structural, which means moral exhortation to the participants is insufficient. Fixing ethics washing requires fixing the structural conditions—the composition of the governing body, the scope of the mandate, the enforceability of the conclusions.

Legitimacy as the product. The output of a washing process is not an ineffective ethics document but a very effective legitimacy document. It tells the public that the technology has been reviewed by experts and found trustworthy, it gives politicians cover for not legislating, and it allows companies to point to the document when challenged. The ethical content is a means to the legitimacy end, not the purpose. This is why ethics washing is worse than no ethics process at all: it consumes the political energy that would otherwise produce genuine constraint.

The scientists cannot delegate. Metzinger’s hardest conclusion from the EU experience is that the researchers who understand these technologies cannot outsource ethical responsibility to governance bodies that have already shown they will trade their red lines for investment. The responsibility is inalienable—it stays with the people who build the systems and the people who understand what those systems might become. This puts a direct obligation on technical researchers that the existence of an ethics committee does not dissolve. It is a specifically demanding form of professional responsibility, requiring researchers to be ethicists of their own work.

Debates & Critiques

The primary challenge to Metzinger’s ethics washing claim is that it proves too much: any ethics governance process will involve the regulated industry, and the claim that this produces washing effectively rules out any legitimate industry participation in ethics. The counter-argument is that Metzinger is not against industry participation but against industry domination—the EU HLEG had a ratio of industry representatives to ethicists that made genuine constraint structurally impossible. A second challenge holds that the EU guidelines, however imperfect, still represent a meaningful baseline that has influenced subsequent legislation, and that dismissing them as washing ignores the real-world effects of imperfect governance. Metzinger’s response is that real-world effects that consist primarily of legitimating continued development, while forestalling the legislation that would impose genuine constraint, are negative rather than positive effects. The deepest challenge is practical: if ethics bodies are always captured, what is the alternative? Metzinger’s answer—that researchers themselves must maintain the red lines—has no clear enforcement mechanism and depends on a degree of professional solidarity that competitive markets tend to erode. The problem he names is real; the solution he proposes is demanding to the point of seeming aspirational.

In the [YOU] on AI Field Guide

Origin

Key Ideas

Debates & Critiques

Related Entries

Further Reading