Ethics Washing, and How to Actually Spot It

Ethics washing, and how to actually spot it.

A set of AI principles costs an afternoon and a designer. A mandate that can actually stop a launch costs a budget line, a reporting relationship, and someone willing to push forward. The gap between those two is the entire phenomenon.

“Ethics Washing” gets discussed as if it is the language: too many adjectives, too much talk of “responsible AI” with too little underneath. That is a real pattern, but it is also the easy half of the problem, because language is exactly what a genuinely serious programme uses too. The harder, more useful question is not how a page reads. It is whether three specific things exist behind it, none of which show up in the language at all.

The three things that actually distinguish it

The first is a mandate: does anyone have the explicit, written authority to say no to a launch, and does that authority sit high enough in the organisation to survive a disagreement with a manager under deadline pressure. A principles document with no one empowered to enforce it is not a mandate, however sincerely it was drafted.

The second is an escalation path: when someone inside the organisation raises a concern, is there a defined route it travels, with a defined person accountable for answering it, or does it simply enter a general inbox and depend on whoever happens to read it that week? An escalation path that exists informally, through whoever you happen to know, is not a path. It is luck, and luck is not something a regulator, or an affected person, can rely on.

The third is a budget line: has anyone actually been given the headcount, the tooling, and the time to do this work, or is “ethics” one more responsibility folded into a role that already had a full-time job description before AI arrived. An organisation that assigns this to someone’s fifth priority has made a real decision about how seriously it takes the fourth, third, second, and first ones instead.

Why the language test fails on its own

Relying on tone alone produces false positives and false negatives in both directions.

• A programme can be genuinely serious and still write in dry, unremarkable, faintly bureaucratic language, because the people running it are spending their energy on the mandate -> rather than the page.
• A programme can also be almost entirely decorative and still read as careful and specific, because writing something that sounds rigorous is a much smaller task -> than building the structure the words describe.

Sincerity is not visible from outside, but a reporting line, a budget, and a documented escalation path are: they can be asked about, named, and checked, which is exactly why they are the more useful test. It’s the same distinction our own methodology draws when rating a credential on the map: what’s claimed, versus what can actually be checked.

What to actually ask

In practice this means asking three blunt, checkable questions rather than reading the principles page at all.

• Who, by name or by role, can stop a launch on ethics grounds, and have they ever actually done it.
• Where does a raised concern go, in writing, and who is accountable for answering it within a stated timeframe.
• What is this year’s budget and headcount for the function.

An organisation that can answer all three specifically, is very likely doing real work. One that can only point back to the page -> is the pattern this piece is named for.

Written by us at Ethics Directive. If anything here needs correcting, we will say so in the open, dated.