Welcome. This module turns from broad ethical principles to a more difficult analytical question: when does corporate communication about “responsible” or “ethical” AI merely describe governance, and when might it help legitimate practices that remain ethically inadequate?
For a thesis on corporate communication, the key discipline is methodological restraint. A public statement rarely lets us know what executives “really intended.” But it can still be analysed rigorously for what it claims, what evidence it supplies, which harms it foregrounds or sidelines, and what the communication is likely to make appear reasonable or settled. By the end of this lesson, you will be able to construct criteria that identify an ethics-washing risk without treating suspicion as proof of corporate deception.
From accusation to assessable claim
“Ethics-washing” is often used as a quick accusation: a company talks about ethics to protect profits, avoid regulation, or improve its image while changing little. That may sometimes be a plausible interpretation, but it bundles together several distinct claims:
- A claim about communication: ethical language is prominent, reassuring, or vague.
- A claim about governance: the organisation lacks meaningful procedures, authority, accountability, or remedies.
- A claim about effects: communication may reduce scrutiny, narrow debate, or make contested practices seem acceptable.
- A claim about intention: the organisation knowingly uses ethics language to mislead.
Only the fourth is a direct claim about corporate motives. It is also the hardest to establish from a corpus of public communication. A company could publish a vague ethics statement because of calculated image management, but also because its governance is immature, because information is commercially sensitive, because teams disagree, or because a general statement is aimed at multiple audiences. None of these explanations automatically make the statement adequate—but they do matter for what you can credibly conclude.
For your project, it is better to treat ethics-washing as an analytical interpretation of the relationship between ethical claims, governance evidence, and communicative effects, rather than a diagnosis of hidden intent.
A defensible conclusion therefore sounds like:
This communication shows a high risk of ethics-washing because it makes expansive ethical claims while providing little accessible evidence of independent oversight, enforceable constraints, affected-party participation, or mechanisms for remedy. Its framing is consistent with a legitimating function, although the analysis cannot establish corporate intent.
This language does three useful things. It identifies observable grounds, makes a bounded interpretive claim, and preserves an appropriate evidentiary limit.
The image makes a vital distinction. Core ethical principles are not implementation mechanisms. A company can endorse fairness, transparency, privacy, and human autonomy without explaining who checks compliance, what happens when harm occurs, or whether affected people can challenge decisions. The gap between those two layers is not proof of ethics-washing, but it is precisely where critical assessment should begin.
Why principles alone are insufficient evidence
Ethical principles matter: they can articulate commitments, guide internal discussion, give stakeholders vocabulary for challenge, and set expectations that later become enforceable. The problem arises when an organisation treats principles as though stating them were equivalent to meeting them.
Read the following two sections from Ethics as a Service: A Pragmatic Operationalisation of AI Ethics. They provide a useful basis for moving from “Does the company have principles?” to “How are those principles translated, contested, monitored, and revised?”
Ethics as a Service: A Pragmatic Operationalisation of AI Ethics
Read this article’s discussion of the limits of principle-based ethics and its alternative: ongoing, publicly defensible ethical governance. It offers concrete grounds for assessing whether an ethics claim is embedded in practice or reduced to compliance theatre.
Begin with the section “Limits of Principlism and Translational Tools.” Read the discussion of tool limits and tick box ethics, paying special attention to the distinction between validation, verification, and evaluation. Then move to the following section, beginning “With this in mind,” and read the proposed process model. Focus on why repeatable procedures, stakeholder involvement, transparent tradeoffs, and oversight matter more than an abstract list of values.
The article identifies a common failure mode: a company can select a fairness, explainability, or risk tool that fits its preferred definition of the problem. A narrowly technical “fix” may then obscure a socially produced harm. For example, improving demographic accuracy in a facial-recognition system does not answer the prior question of whether facial recognition should be used in a particular policing, employment, welfare, or border-control setting.
This gives you an important methodological rule:
Do not evaluate an ethical claim only by whether the stated principle is present. Evaluate whether the claim identifies the relevant decision, the trade-offs, the authority to intervene, and the process by which the commitment is revisited after deployment.
In UX terms, a published principle is closer to a service promise than to evidence that the service works. A company saying “we respect user autonomy” is not enough. An assessment asks what user choices actually exist, whether they are intelligible and non-coercive, whether refusal is possible without excessive penalty, and how harmful outcomes are corrected.
A cautious assessment framework
The following six criteria can become the backbone of an ethics-washing assessment rubric. They do not produce a binary verdict. Instead, they help you describe a communication artifact as showing low, moderate, or high concern, with accompanying evidence and uncertainty.
1. Claim specificity and ethical boundaries
Start by asking whether the ethical claim is meaningful enough to evaluate.
A weak statement says:
We are committed to safe, fair, and responsible AI.
A stronger statement identifies:
- the system or product covered;
- the affected groups and relevant context of use;
- the harm or risk being addressed;
- the company’s concrete commitment;
- limitations, exclusions, or prohibited uses;
- the decision-maker responsible for implementation.
Ethics language becomes difficult to assess when it uses broad positive terms—“trust,” “human-centred,” “ethical,” “inclusive,” “safe”—without defining their operational meaning. Such abstraction can create a reassuring moral atmosphere while leaving the company discretion to decide what the terms mean in each case.
Cautious indicator of concern: expansive ethical vocabulary with no scope, thresholds, definitions, trade-offs, or red lines.
Counter-evidence: clear statements of where a product should not be used, what risks cannot be fully eliminated, and which decisions remain contestable. Acknowledged limitations often make a claim more credible, not less.
2. Material connection to the AI lifecycle
An ethical claim is more substantive when it is connected to decisions that shape a system throughout its lifecycle. The relevant question is not “Does the company conduct an assessment?” but “At what point, with what consequence, and how often?”
Look for evidence across at least three moments:
| Moment | What to look for |
|---|---|
| Before development or release | Impact assessment, stakeholder consultation, documented risk acceptance, ability to halt or redesign a project |
| During development | Data governance, safety testing, red teaming, accessibility or fairness evaluation, documented trade-offs |
| After deployment | Monitoring, incident reporting, external feedback channels, model or policy revision, suspension or withdrawal mechanisms |
A one-time ethics review can be useful, but it is not sufficient evidence of continuous governance. AI systems, deployment environments, user practices, and social consequences change over time. An organisation that describes ethics solely as a launch-stage checklist may be treating it as a reputational or compliance milestone rather than an ongoing responsibility.
Cautious indicator of concern: ethics is communicated as a certification-like achievement—“reviewed,” “approved,” “principles applied”—without explaining ongoing monitoring or what happens when problems are found.
Counter-evidence: documented processes for post-deployment evaluation, incident response, revision, and cessation of harmful uses.
3. Authority, independence, and enforceability
A company may publicise an ethics board, advisory council, safety team, or review process. Do not assume that its existence is either meaningless or effective. Ask what institutional power it actually has.
Key questions include:
- Who appoints members and controls the budget?
- Can the body review high-value or strategically important products?
- Can it delay, modify, or stop deployment?
- Are decisions binding, advisory, or unknown?
- Does it report publicly, including disagreements and limitations?
- Can employees raise concerns without retaliation?
- Is there independent review or only internal assurance?
This criterion is especially important in B2G or defence-oriented contexts, where commercial confidentiality, security constraints, and government procurement can restrict disclosure. Restricted disclosure does not automatically invalidate a governance claim. However, when information cannot be public, a robust claim should identify alternative accountability arrangements: independent reviewers with access, legislative oversight, procurement safeguards, inspection powers, audit trails, or a credible redress route.
Cautious indicator of concern: a highly visible governance body with unspecified mandate, authority, independence, or outcomes.
Counter-evidence: clear decision rights, documented escalation paths, transparent reporting, protection for dissent, and accountability that is not solely controlled by the business unit benefiting from deployment.
4. Affected-party participation, justice, and remedy
Corporate ethics communication often centres the company’s engineers, executives, and experts. A more capacious account asks whose lives, work, rights, or opportunities are affected—and whether those people can influence decisions.
Assess whether the text recognises:
- direct users;
- people subject to AI-mediated decisions but not choosing the system;
- workers who label data, moderate content, or operate systems;
- communities exposed to surveillance, environmental costs, or public-sector automation;
- groups historically exposed to discrimination or unequal error rates;
- institutions responsible for public accountability.
The image’s “citizen participation” and “effective redress” components are useful here. Participation is not satisfied merely by a company saying it “listens to feedback.” Look for who was included, at what stage, with what influence, and whether disagreement affected the decision. Similarly, redress requires more than a support email. It involves a way to contest a decision, seek review, correct data, obtain explanation where appropriate, and receive a remedy proportionate to the harm.
Cautious indicator of concern: ethical narratives focus on product benefits and user choice while omitting non-users, affected communities, workers, or people with limited ability to refuse the system.
Counter-evidence: specific accounts of stakeholder engagement, mechanisms for contestation and remedy, and evidence that participation changed design, deployment, or policy.
5. Transparency matched by accessible evidence
Transparency is a particularly flexible corporate term. It may refer to publishing principles, describing a model’s capabilities, giving users explanations, issuing a transparency report, or disclosing safety evaluations. These are not equivalent.
Analyse transparency as a relationship between claim and evidence:
| Corporate claim | Evidence that would make it assessable |
|---|---|
| “Our AI is fair” | Evaluation scope, relevant populations, metrics, known limitations, mitigation decisions, monitoring plan |
| “We use rigorous safety testing” | Test types, risk categories, thresholds, independent challenge or review, findings at an appropriate level of detail |
| “We are accountable” | Named roles, governance authority, reporting routes, investigation process, consequences, remedies |
| “We protect privacy” | Data categories, purpose limits, retention approach, user controls, third-party access rules, audit or compliance evidence |
Commercial sensitivity, security, privacy, and misuse risks can legitimately limit disclosure. Your criterion should therefore not demand total openness. Instead, ask whether the company provides enough information for an informed external audience to evaluate the claim, or whether it explains why information is withheld and what independent assurance substitutes for disclosure.
Cautious indicator of concern: “transparent” is asserted while criteria, methodologies, limitations, outcomes, or responsible actors are inaccessible.
Counter-evidence: meaningful disclosure, proportionate explanations of legitimate non-disclosure, and credible external or independent scrutiny.
6. Discursive effects: displacement and the legitimacy buffer
The most distinctive critical criterion concerns not only what a company does, but what ethics communication does in public discourse. In the ethics-washing literature, ethical language may work performatively: it can make a controversial practice appear settled, responsible, and already governed. This may reduce pressure for regulation or redirect attention from structural questions to manageable technical improvements.
The central question is:
Does the ethical narrative broaden the space for contestation, or does it narrow the problem so that the company’s preferred solution appears to be the only reasonable one?
For example, a company may frame an AI surveillance system as a matter of “accuracy and bias reduction.” That framing can be ethically relevant. But it can also displace questions about necessity, consent, discriminatory targeting, public-sector power, democratic control, and whether the system ought to be deployed at all.
Read this conceptual account of the problem. It will help you distinguish a critique of ethics-washing from an indiscriminate rejection of all corporate ethical work.
This ACM paper develops a nuanced critique of corporate ethics work. Read it to understand three different concerns: limited social impact, constrained ethical scope, and the performative possibility that ethics rhetoric creates a legitimacy buffer.
In the section “Three Critiques of Ethics Washing from Within Moral Philosophy,” read the three-part critique. Focus on the distinction between instrumental value, intrinsic ethical commitment, and the public effects of ethics rhetoric. For context on the “legitimacy buffer,” also read the latter half of “What’s Wrong with Ethics Washing,” beginning the discussion of self regulation and scrutiny. Notice that the author explicitly allows for positive impacts and fact-specific exceptions rather than treating all corporate self-regulation as inherently deceptive.
This criterion must be used particularly carefully. You normally cannot demonstrate that a statement actually changed regulation or public opinion from the statement alone. What you can show is that the communication:
- frames a systemic or political problem as a technical optimisation problem;
- presents internal self-regulation as an alternative to external accountability;
- foregrounds a visible initiative while backgrounding business practices central to the harm;
- depicts a contested deployment as inevitable, then limits ethical debate to implementation details;
- omits alternatives such as non-deployment, stronger regulation, or redistribution of decision-making power.
These are signs that a message may function as a legitimacy buffer. They are not proof that the organisation intended to manipulate its audience.
A practical rubric: evidence, not mind-reading
For each communication artifact, create a short assessment record. The unit could be a blog post, policy page, safety report, product announcement, executive speech, or procurement-facing document, depending on your later sampling decisions.
Use a four-part structure:
| Field | What you record |
|---|---|
| Claim | The precise ethical, safety, transparency, or responsibility claim, quoted with context |
| Evidence in the artifact | Mechanisms, definitions, scope, outcomes, limits, governance details, and links supplied by the company |
| Omissions or tensions | Missing stakeholders, unclear authority, absent remedy, mismatch between broad claim and narrow evidence, displaced alternatives |
| Cautious interpretation | Low, moderate, or high ethics-washing concern; alternative explanations; evidence needed to revise the assessment |
A coding note might look like this:
Claim: “Our AI safety framework ensures responsible deployment.”
Evidence in the artifact: It names three safety principles and says teams conduct pre-release testing.
Omissions or tensions: The framework does not define “responsible,” disclose thresholds for release, identify who can override commercial pressure, mention post-deployment monitoring, or provide a channel for affected parties to seek remedy.
Interpretation: Moderate-to-high concern. The claim is broader than the governance evidence made accessible in this artifact. The framing is consistent with responsibility signalling and potentially with ethics-washing, but intent cannot be inferred. A linked technical report, governance charter, incident process, or independent audit could alter this assessment.
This format prevents two common errors.
The first is intentionalism: “They are ethics-washing because they want good publicity.” Unless you have internal evidence, treat that as unverified speculation.
The second is naive verification: “They published an ethics board and safety policy, so the commitment is substantive.” Public documentation is relevant evidence, but governance quality depends on authority, scope, independence, implementation, and outcomes.
A third possibility is often most realistic: mixed practice. A company may invest seriously in testing and internal governance while also communicating those efforts strategically to build trust, differentiate itself, reassure regulators, or protect its reputation. Responsibility signalling and reputation management are not automatically unethical. Your task is to assess whether ethical communication is proportionate to, and supported by, meaningful accountability, rather than to assume that strategic communication invalidates all ethical work.
Calibrating conclusions
To make your thesis analytically credible, match the strength of your conclusion to the evidence available.
| Evidence available | Appropriate conclusion |
|---|---|
| One public statement with generic ethics language | The claim is vague or difficult to assess; further evidence is required |
| Several texts repeat broad commitments but lack operational detail | The communication pattern indicates a gap between ethical rhetoric and publicly accessible governance evidence |
| Strong rhetoric conflicts with documented practices, external investigations, or the company’s own disclosed outcomes | The pattern provides stronger grounds to interpret communication as decoupled from practice and consistent with ethics-washing |
| Independent oversight, enforceable rules, ongoing reporting, participation, and remedy are documented | The communication is more strongly supported by substantive governance, though limitations and power asymmetries may remain |
| Evidence is incomplete or contradictory | Classify the case as indeterminate or mixed; specify what cannot be concluded |
The last row is important. “Indeterminate” is not a weak result; it is often the most honest one. Corporate materials are designed for communication, not necessarily for independent evaluation. A rigorous analysis makes visible both what is present and what cannot be known from the corpus.
When writing up a finding, distinguish three levels explicitly:
- Description: “The company frames safety as a technical testing process.”
- Interpretation: “This narrows responsibility toward model behaviour and away from downstream institutional uses.”
- Evaluation: “Because no governance authority, stakeholder participation, or redress is disclosed, the statement has a moderate ethics-washing risk.”
This distinction will later protect your analysis from overclaiming when comparing OpenAI, Anthropic, and Microsoft across consumer, enterprise, and government-facing contexts.
Key takeaways
A cautious ethics-washing assessment does not attempt to read corporate minds. It evaluates the relationship between ethical rhetoric, visible governance mechanisms, omissions, and plausible legitimating effects.
The most useful criteria ask whether communication provides:
- specific, bounded, assessable ethical claims;
- evidence that commitments shape decisions across the AI lifecycle;
- independent and enforceable governance authority;
- participation, justice, contestability, and effective remedy;
- transparency supported by accessible, proportionate evidence;
- an ethical framing that keeps structural alternatives open rather than creating a legitimacy buffer.
Use these criteria as risk indicators and comparative lenses, not as a shortcut to calling a company deceptive. The strongest conclusion is usually conditional: a message may be consistent with ethics-washing, may risk functioning as a legitimacy device, or may show a gap between rhetoric and accessible evidence.
In the next lesson, you will apply this distinction more directly by separating ethics-washing, responsibility signalling, and reputation management as different analytical interpretations of corporate AI communication.
Can't find a good explanation? Sign up and we'll make it for you
Sign up