AI Ethics Boards That Never Blocked a Project

I kept thinking about the Google Advanced Technology External Advisory Council after it died. It was announced in March 2019 with great fanfare: a panel of experts who would review Google's AI projects for ethical concerns. It had a charter. It had a press release. It had a name so bureaucratic it sounded permanent. Ten days later it was dissolved because of controversy over who was on it. The thing that stayed with me was not the dissolution. It was that in its brief existence, the council never reviewed a single project. The structure existed. The practice never started.

Meyer and Rowan published their paper on institutionalized organizations in 1977, forty-two years before that press release. Their argument was that formal organizational structures are often designed for legitimacy rather than efficiency. Organizations adopt structures that signal conformity to institutional expectations: mission statements that say the right things, committees that look like oversight, processes that resemble governance. The structures are ceremonial. Their purpose is to say to the external world that the organization is doing what organizations in its field are expected to do. Meyer and Rowan called the gap between the formal structure and the actual activity decoupling, and they argued that decoupling is not a bug. It is the point. The formal structure handles legitimacy. The actual activity handles the work. Keeping them separate protects both.

When I read Robey and Boudreau (1999) who cite Meyer and Rowan directly, the language landed differently. They say organizations "tend to conform to institutional models while resisting attempts at reform, even where organizational efficiency is threatened." That is a precise description of what happens when an AI ethics board reviews a product, the product team has a revenue target, and the director of product says, thank you for your input, and launches anyway. The ethics board is the institutional model. The revenue target is the actual model. The decoupling means the product ships with whichever flagged issues survive the revenue conversation.

The Google council was an extreme case because it never got past the ceremonial stage. Most AI ethics boards at major technology companies are less dramatic. They have regular meetings. They review projects. They produce recommendations. And the pattern is the same across the industry: the recommendations are almost always advisory, the product launches almost always proceed, and the board exists on paper as evidence that the company takes AI ethics seriously. Microsoft published its AI principles in 2018 and established an internal AI ethics committee to review sensitive uses. The same year, Microsoft bid on the Pentagon's Joint Enterprise Defense Infrastructure contract, worth up to ten billion dollars, for cloud services supporting military operations including AI-enabled capabilities. Microsoft employees publicly protested. The AI ethics committee had been formed, had published principles about fairness, reliability, privacy, inclusiveness, transparency, and accountability, and the JEDI bid proceeded. The structure said the company would weigh ethical implications before pursuing sensitive AI contracts. The activity said revenue and strategic positioning override ethical deliberation. That is decoupling.

Ethics washing has become the term for this pattern. It means adopting the trappings of ethical governance without subjecting decisions to genuine ethical constraint. You hire an AI ethics officer. You publish a responsible AI framework. You form an advisory board. You attend conferences on trustworthy AI. And the product that your engineers warned would cause measurable harm to vulnerable populations ships on schedule because the ethics function reports to the legal department and up the chain to the CEO who reports to shareholders who did not invest in an ethics board; they invested in revenue growth.

I think AI ethics boards are the most pure example of institutional decoupling I have seen in the technology industry, and I think the reason is structural, not malicious. The ethics board is an institutional response to institutional pressure. Regulators are signaling that AI governance matters. Customers are asking about responsible AI. Industry bodies are publishing standards. The normative pillar that Scott (1995) described, the professional expectations that make certain practices feel mandatory, is building rapidly. Companies respond by creating the visible structures that the institutional environment demands. An AI ethics board is a legitimate structure because other legitimate organizations also have one. The CEO can say in an earnings call that the company has a responsible AI framework, and the sentence works because the listener knows what that means. The decoupling happens because the economic logic and the institutional logic are different systems. The institutional logic demands an ethics board. The economic logic demands that profitable products ship. The board is designed to satisfy the first demand. The launch pipeline is designed to satisfy the second. Meyer and Rowan would say the whole arrangement works precisely because the board and the launch pipeline are loosely coupled. If they were tightly coupled, the board could stop launches, and the company would lose revenue and competitive position. Loose coupling lets the company have both: the legitimacy of governance and the flexibility of unconstrained product decisions.

The ethics washing critique is real but it is not the full story. What looks like ethics washing from the outside is often the predictable outcome of an organizational structure that was never designed to constrain. The board members I know who sit on AI ethics committees are genuinely thoughtful people who take their role seriously. They flag risks. They write detailed assessments. And in many cases they make a difference around the edges, on projects where the revenue is small enough that the committee's recommendation can be accepted without a fight. The moment a recommendation threatens a major product launch or a strategic partnership, the coupling proves loose enough to absorb the recommendation without changing the decision. The board operates in the zone where ethics does not cost too much. That zone shrinks as the revenue at stake grows.

DiMaggio and Powell (1983) identified three pressures that push organizations toward convergence. Mimetic pressure is the strongest driver of AI ethics boards. When a major technology company forms an AI ethics committee, the uncertainty about what responsible AI governance should look like is high. The safest response is to copy what peers are doing. Every AI ethics board at every major technology company is shaped by the same conference panels, the same consulting reports, the same job postings for the same niche roles. The boards converge in structure because the organizations are all imitating the same models under the same uncertainty. The result is a field-level isomorphism of AI governance structures, with the same frameworks, the same principles, the same board compositions, and the same decoupling from operational decisions.

I wrote about this convergence pattern before in my post on everyone copying everyone else's AI strategy. The institutional pressure that made companies copy each other's AI adoption is the same pressure driving the ceremonial adoption of AI ethics boards. The same mimetic logic that gave every SaaS vendor a chatbot also gave every large technology company an AI ethics page and a board charter. The isomorphism covers both the product and the governance of the product, which means the governance is as copied as the technology it is supposed to govern.

I do not think this is permanent. Meyer and Rowan argued that decoupling is unstable because external scrutiny eventually exposes the gap between formal structure and actual practice. When an AI ethics board issues a public recommendation against a project and the project proceeds anyway, the ceremony stops working. The legitimacy function collapses because the public can see the decoupling. The structure that was designed to signal ethical governance starts signaling the opposite: that ethics is a cover for business as usual. Some organizations will respond by tightening the coupling, giving the board actual authority to block projects, embedding ethical review into the product cycle rather than layering it on as an advisory afterthought. Those organizations will be the exception, because tight coupling creates genuine constraint and genuine constraint is expensive. The organizations that do it will do it not because institutional pressure demands it but because their leadership believes ethical boundaries are a strategic asset.

I want to close on what this means for the IS field. We study technology adoption in organizations. We have frameworks that explain adoption patterns, and institutional theory is one of the sharpest tools for understanding why organizations converge on similar structures. But the same convergence that produces broad adoption also produces ceremonial adoption, structures that exist for legitimacy rather than to improve outcomes. The AI ethics board is the textbook example. The research question I keep turning over is whether ceremonial adoption has a half-life: whether decoupled structures eventually collapse under scrutiny or whether organizations can maintain the gap indefinitely by managing external perception. I am not sure about this yet. The answer probably depends on how much external scrutiny the organization faces and how willing the public is to accept the ceremonial substitute. But I am watching the AI ethics boards to find out.