The independence problem

Give someone the title without the structure, and you have not hired an ethicist. You have hired someone whose job is to feel responsible when something goes wrong.

It rarely fails loudly

Most AI ethicist roles fail quietly, not in one dramatic incident. A launch date that cannot move, a reporting line that runs straight into the team whose product depends on the answer, a review that gets rescheduled until it is no longer relevant. Six months later the person in the role has learned, without anyone saying so, which findings are safe to raise and which are not. That is not a failure of character. It is a predictable response to a role with no actual independence built into it.

There's a second, quieter risk in the same shape. Give someone the title without real authority, and when something does go wrong, the organisation has a ready-made place to point: the person hired to hold the ethics line, rather than the decisions that boxed them in. Independence exists to stop that gap from opening in the first place, not to hand someone a badge to wear while it does. There's a name for this pattern: ethics theatre, the appearance of responsibility without the substance, a board or a badge with nothing underneath it. It's what you get when the title shows up before the structure does.

Six things decide whether it holds

Six structural facts, agreed before day one rather than negotiated in the moment, decide whether a role survives its first hard finding: who the person actually answers to, whether they have a real route to be heard if a finding gets softened on the way up, whether their dissent stays on the record as given rather than revised after the fact, whether raising something unwelcome can quietly cost them the engagement, whether conflicts of interest, and confidentiality, are declared and kept as a matter of course rather than left to memory, and whether the role has a fixed end date so independence gets renewed on purpose instead of eroding by default.

Why the excuses sound reasonable

Nobody sits down and decides to bury a finding. It happens a step at a time, and every step sounds sensible on its own: this is a one-off, just for this launch, not a pattern; the risk is small and probably won't land; the people affected will never know, or wouldn't mind if they did; raising it now just slows down people who are already under enough pressure. Each excuse is individually plausible. Strung together across enough small decisions, they add up to a team that has quietly talked itself out of hearing its own ethicist. None of the six clauses stop anyone thinking this way, that's not something a contract can fix. What they stop is a soft no travelling any further than one conversation before it has to be said out loud, on the record, to someone with the authority to disagree.

Who checks the checker

There's an honest objection worth naming here, because a values-focused reader should be the first to ask it: doesn't a fee paid by the hiring organisation create the same conflict this whole page is warning about? A regulator would not let an auditor grade the exam they were paid to sit. It's a fair comparison, and it is exactly why six clauses exist rather than a simple promise of neutrality. Sentient Staff is paid by the organisation, not the practitioner, so the same pressure that can compromise any fee-for-service arrangement is present here too, and we would rather say so than pretend otherwise. What can actually be controlled is what happens next: independence written into the engagement before it starts, a route to be heard that does not run through the account holding the commercial relationship, and confidentiality that means our own interest in keeping a client happy is never the reason a finding gets softened. That doesn't remove the conflict. It's the structure that stops the conflict from quietly deciding the outcome.

Why this is fixable, and usually isn't fixed

It's fixable cheaply, and it usually isn't fixed because nobody thought to ask. It costs a few clauses in an engagement agreement and a willingness to say them out loud before someone starts. What it is not is automatic. Most organisations hiring their first AI ethicist have never had to think about giving someone a real route to be heard, or protecting them from quietly losing the engagement for saying something unwelcome, because most roles don't need it. This one does, and the gap between knowing that in theory and writing it into a contract is where most of these roles quietly lose their teeth.