Almost every major AI company now has some version of an internal ethics function: a review board, a responsible AI team, a trust and safety council, or some combination of the three. These structures have become common enough that their absence would now be more notable than their presence. What remains far less settled is how much real influence these bodies have over the decisions that matter most — what gets built, what gets shipped, and what gets shelved. The gap between the existence of an ethics board and its actual authority is one of the more persistent and under-examined issues in how AI companies govern themselves.
Why Companies Built These Structures in the First Place
The rise of internal ethics functions was driven by a mix of genuine concern and practical necessity. As AI systems moved from research demos to products used by hundreds of millions of people, the range of things that could go wrong expanded correspondingly: biased outputs, privacy violations, misuse by bad actors, psychological effects on vulnerable users, and the sheer reputational risk of a single high-profile incident. Companies that had previously treated ethical review as an academic afterthought found themselves needing a formal process simply to manage that risk at scale.
There was also an external dimension. Journalists, researchers, and regulators increasingly asked pointed questions about how companies were thinking through the societal implications of their products, and having a named team or board to point to became, in part, a communications necessity. That does not mean these teams were created cynically — many are staffed by people who joined specifically because they believed in the mission and wanted real influence over outcomes. But the initial impetus for building the structure and the eventual effectiveness of that structure are two different things, and conflating them has led to some overly generous assumptions about how much these boards actually accomplish.
The Authority Problem
The central limitation of most internal ethics boards is structural rather than personal: they typically advise rather than decide. A board can flag a concern about a model’s behavior, recommend additional testing, or push back on a proposed release timeline, but in most organizational structures, the final call rests with product and executive leadership, who are also the people most directly accountable for growth, revenue, and competitive positioning. When an ethics recommendation and a business priority point in different directions, the ethics board’s role is frequently to be heard, not necessarily to be obeyed.
This asymmetry becomes especially visible during periods of intense competitive pressure — precisely the moments when careful review matters most. Multiple former members of internal ethics and safety teams at major AI companies have described, in interviews, congressional testimony, and public writing, a recurring pattern: their teams’ influence was strongest when public attention on AI safety was high and weakest when a competitor’s product launch created pressure to ship quickly. That pattern suggests the boards function less as an independent check on the business and more as a pressure valve that expands and contracts with external scrutiny, rather than operating from a stable and durable base of authority.
Some organizations have tried to address this by giving ethics and safety leads a formal veto over specific categories of decisions, or by structuring reporting lines so that safety leadership reports to the board rather than to product executives. These structural changes appear to correlate with more durable influence, though they remain the exception rather than the norm across the industry, and even where they exist, they are frequently limited to a narrow set of high-severity decisions rather than the full range of product and research choices that carry ethical weight.
The Departure Pattern
A recurring and telling signal has been the frequency with which senior members of internal ethics and safety teams leave their organizations, often citing frustration with how much their recommendations actually shaped outcomes. These departures are rarely framed publicly as dramatic resignations in protest — more often they are described in measured terms, citing a mismatch between the role’s stated mandate and its actual influence, or a sense that the position had become primarily about documentation and process rather than substantive decision-making power.
The cumulative effect of these departures across the industry has been a noticeable churn in ethics and safety leadership at several major companies over the past few years, with new hires often arriving to rebuild institutional processes and relationships that their predecessors had spent years establishing. This churn carries a real cost: internal ethics functions depend heavily on relationships across an organization — knowing which product leads are receptive to feedback, which technical teams need extra support to implement a recommendation, and which historical incidents inform current risk assessments. When leadership turns over frequently, that accumulated context is lost, and new teams often find themselves re-litigating debates that were, in principle, already settled.
It would be a mistake to read every departure as evidence of bad faith on the part of the company. Some transitions reflect ordinary career moves, and some ethics leads have left specifically because they achieved what they set out to do and wanted a new challenge. But the pattern, taken as a whole across multiple organizations and multiple years, is difficult to explain purely through individual career choices, and it has become one of the more commonly cited pieces of evidence for skeptics who question how much structural power these boards genuinely hold.
What More Effective Models Look Like
Not every internal ethics function operates the same way, and some structural choices appear to correlate with greater durability and influence. Boards that include members with fixed terms and protections against being removed by the executives whose decisions they are reviewing tend to exhibit more independence than those that serve entirely at leadership’s discretion. Boards with a formal, documented process for escalating disagreements — including the ability to flag unresolved concerns to a company’s board of directors rather than only to product leadership — appear better positioned to have their concerns taken seriously when internal incentives point the other way.
External representation has also emerged as a meaningful, if imperfect, lever. A handful of companies have added outside members — academics, former regulators, civil society representatives — to their ethics boards, on the theory that people without a direct stake in a product’s commercial success are less likely to defer automatically to internal pressure. This model brings its own complications, including questions about how much real information outside members actually receive and how much influence they can exercise without deep, ongoing familiarity with a company’s internal systems. But it represents one of the more promising available correctives to the core authority problem, since it introduces at least some perspective that is not structurally dependent on the company’s short-term performance.
Transparency requirements — publishing not just the existence of an ethics review process but some account of what it actually reviewed, what it recommended, and whether those recommendations were followed — offer another partial solution. Very few companies currently do this in a rigorous way, and the ones that publish the most detailed accounts tend to be smaller organizations without the same competitive pressure faced by frontier labs. Whether that changes as public and regulatory attention on AI governance continues to grow will likely determine whether internal ethics boards evolve into genuinely load-bearing parts of company decision-making, or remain a largely advisory function whose influence rises and falls with the news cycle.
The Broader Stakes
The question of whether internal ethics structures actually shape outcomes matters well beyond any single company. As AI systems take on more consequential roles — in hiring, lending, healthcare, and public services — the gap between an ethics board’s stated mandate and its actual authority becomes a gap in real-world accountability, not just an internal governance curiosity. Users, regulators, and the public generally have no direct visibility into how these internal processes function, and are left inferring their effectiveness from indirect signals: departures, public statements, and the rare cases where internal disagreements become visible through leaks or litigation.
Building internal ethics functions with genuine authority — durable reporting structures, external accountability, and real transparency about outcomes — is not a purely altruistic exercise for AI companies. It is also, increasingly, a competitive and reputational necessity, as public trust becomes a meaningful factor in enterprise adoption and regulatory goodwill. Whether that incentive proves strong enough to close the authority gap that has defined these structures so far remains one of the more consequential open questions in how the AI industry governs itself.
What Employees Inside These Companies Say Would Actually Help
Interviews and public writing from people who have served on or worked closely with internal ethics functions point to a fairly consistent list of changes they believe would make these structures more effective, independent of any single company’s specific circumstances. Near the top of that list is simply timing: ethics and safety review that happens early enough in a product’s development to actually influence design decisions, rather than being scheduled as a final check shortly before launch, when the cost of substantially reworking a system has already become prohibitively high and the incentive to find reasons to proceed anyway is correspondingly stronger.
A second recurring theme is resourcing that scales with a company’s growth rather than staying fixed while user bases and model capabilities expand. Several former ethics staffers have noted that their teams remained roughly the same size for years even as the products they were reviewing grew dramatically in scope, reach, and complexity — a mismatch that left review processes increasingly superficial by necessity, simply because there were not enough people to examine every consequential decision with real depth.
A third theme, somewhat less commonly voiced but no less significant, is a desire for internal ethics work to be measured and rewarded using metrics that do not simply mirror the business metrics used to evaluate product teams. When an ethics function’s own performance reviews are implicitly tied to how smoothly it enabled launches to proceed on schedule, the incentive to flag serious problems is quietly undermined by the same career pressures that affect every other part of the organization. Solving that particular problem may ultimately require rethinking not just the structure of ethics boards, but how the people who staff them are evaluated and promoted within the broader company — a change that goes well beyond adding a board or a policy document, and one that few organizations have attempted in any comprehensive way so far.