A company launches a product. The press release mentions “human-centered design” four times. The landing page features a photograph of someone’s hands at a workbench. The CEO’s LinkedIn post promises that “people always come first.” Three months later, the same company announces a round of layoffs executed through an automated HR platform that sent termination emails before managers were notified. Nobody blinks.

cover

This is human washing, the art of deploying the rhetoric of human involvement, human care, and human accountability as a trust signal, while the organizational reality underneath stays exactly as it was. It is not a new phenomenon, but it is becoming systematic, and it deserves the same critical scrutiny we have learned to apply to its older cousin, greenwashing.

Defining the problem

The term borrows its structure directly from greenwashing, the practice of misleading stakeholders about environmental performance to exploit the growing premium placed on sustainability. The analogy is precise. As Moviweb’s analysis of the human washing phenomenon puts it, the mechanism emerges exactly where humanity is no longer taken for granted in communication, and for this very reason becomes a value to declare, defend, and demonstrate. When automation becomes the norm, claiming the human becomes a differentiator. And differentiators get marketed, stretched, and eventually emptied.

The academic scaffolding for this type of analysis already exists. A 2022 paper published on SSRN, From Greenwashing to Machinewashing, proposes a conceptual mapping from greenwashing to “machinewashing” (defined as misleading communication about ethical AI systems) using a reasoning-by-analogy approach. The same methodology applies cleanly to human washing: take the structural model of greenwashing, substitute “environmental responsibility” with “human involvement,” and you get a ready-made framework for identifying, categorizing, and measuring the deception.

Human washing, in operational terms, is a strategy an organization adopts to engage in misleading behavior (in communication or in action) about the actual role of human judgment, human labor, or human accountability in its processes and products. It involves misleading information communicated via language, visuals, certifications, or the design of the systems themselves. In the worst cases, like machinewashing’s use for lobbying, it is also deployed to forestall stricter regulation by projecting an image of voluntary responsibility that makes oversight seem redundant.

Why it works: the trust economy and the AI commodification problem

The deeper cause is structural. Generative AI has lowered the cost of producing competent-looking content, decisions, and products to near zero. As the Moviweb piece observes, AI has made creativity easily replicable and lowered the threshold of access to “good form”, generating a background noise of correct, well-constructed, but interchangeable output. When everything looks smooth and optimized, the audience stops believing the message was made for them. Declaring the human becomes a way to restore emotional density to communication that has been bleached of it.

This creates a commercially rational incentive to overclaim. Companies that can credibly signal “human involvement” command higher trust, higher prices, and better talent. The problem is that “credibly” and “actually” are not the same thing, and in the absence of any standardized verification, the claim costs almost nothing to make. We are already living through an analogous dynamic in AI ethics: the machinewashing literature documents how organizations adopt “AI ethics” frameworks as a form of symbolic action, including covert lobbying to prevent stricter regulation, rather than as a substantive commitment. Human washing operates on the same incentive structure, at the layer of labor and responsibility rather than algorithm and model governance.

There is also a regulatory angle worth tracking. The EU’s AI Act, NIS2, and DORA all include provisions touching on human oversight and accountability. A company that can point to a “human-in-the-loop” policy, however hollow, has a ready answer to auditors, regulators, and incident investigators. The claim performs compliance theater before it performs actual compliance.

The seven signs: a checklist adapted from greenwashing taxonomy

The greenwashing literature has spent two decades developing practical tools for identifying misleading environmental claims. A comprehensive review published in the Journal of Consumer Affairs catalogues recurring patterns (“sins”) that can be adapted directly to human washing. Below is a working checklist.

1. Vagueness. “People-first,” “human-centered,” “handcrafted at scale”: no metrics, no definitions, no audit. If the claim cannot be measured, it should not be trusted.

2. Irrelevance. Calling something “human” when human involvement is non-material to the concern at hand. A hiring process can be described as “human-reviewed” if one person skims an AI-generated shortlist for thirty seconds. Technically true. Operationally meaningless.

3. Lack of proof. No third-party audit, no published methodology, no verifiable data. “We care deeply about our people” without payroll transparency, turnover metrics, or worker survey results is decoration, not evidence.

4. False framing. Showing the artisan’s hands in the product video while the supply chain runs on precarious platform labor. The visible “human” is real; its representativeness is zero.

5. Worshipping the label. Saturation of human imagery and language across brand assets while the process remains automated or outsourced. The density of the claim is inversely proportional to its substance.

6. Hidden trade-offs. Emphasizing human touch in one part of the value chain while obscuring automation, outsourcing, or exploitative conditions in another. The “human-curated” playlist was assembled by a model; the actual humans who recommended it were contractors paid per item.

7. The responsibility gap. In AI-adjacent contexts: claiming “human-in-the-loop” without specifying who, with what authority, at what point in the decision, with what override capability, and with what accountability when things go wrong. The How organizations can adopt AI security tools without losing control piece on this blog notes that effective human oversight requires clear accountability structures, traceability of AI decision-making processes, and defined roles, not a badge on a press release.

Three cases worth examining closely

Employer branding campaigns. The pattern is by now a genre: behind-the-scenes photos of open offices, testimonials from employees who look like they are having the time of their lives, a careers page that leads with “we are a family.” Cross this with publicly available data (Glassdoor reviews, LinkedIn headcount trends, employer litigation records, voluntary attrition rates) and the gap between narrative and reality is often significant. The culture-fit trap dissected in a recent post on this blog is closely related: “culture fit” rhetoric is itself a form of human washing, projecting a warm, values-driven hiring process onto what is often intuition, homophily, and institutional risk-aversion. The human claim obscures the structural filter.

“Human-curated” AI products. A content platform, a cybersecurity threat intelligence feed, or a due-diligence service can now routinely label their output “human-curated” or “expert-verified.” The practical question is: what does that mean in operational terms? How many analysts? At what stage? With what override authority? What is their error rate? In most cases, these questions are not answerable from public materials, because no public materials exist. The label is doing all the work. This is especially acute in security, where “human-in-the-loop” is increasingly a regulatory expectation: a policy that says so without the governance architecture to back it up is machinewashing wearing a human mask.

Corporate social responsibility and human rights supply chain claims. Brands that market handmade, artisanal, or ethically sourced products regularly face scrutiny that reveals the gap between claim and supply chain reality. The same logic applies to tech: a company that claims to employ only fairly compensated, permanently contracted human workers for content moderation, data labeling, or security operations, while actually routing that work through sub-contractors in lower-cost jurisdictions with high turnover, is running a version of the same playbook. The human is present. The conditions that make the claim meaningful are not.

Practical questions for anyone who wants to verify

Documents to request: HR policies with specific provisions on outsourcing and contractor classification; aggregate payroll data or compensation band disclosures; third-party audit reports on labor practices; model governance documentation for AI systems claiming human oversight (decision logs, override rates, incident reports); attrition rates by role and level.

Interview targets: line managers (not comms), workers in named roles, vendor contract managers, the MLOps or AI governance function if one exists. Comms teams are not sources on this topic; they are the phenomenon.

For AI-adjacent claims specifically, the machinewashing framework cited above proposes examining communication, action, and the underlying algorithm as three distinct layers where misleading signals can be embedded. A “human-in-the-loop” claim can be technically present in the documentation while being architecturally absent from the system. Ask for the decision logs.

For policy makers, the greenwashing prevention literature, including recent academic work on ESG anti-greenwashing provisions, points toward mandatory disclosure, standardized metrics, and independent attestation as the only reliable remedies. The same model applies here: voluntary human-washing claims should face the same pressure for third-party verification and structured disclosure as environmental claims now do in the EU.

If “human” has become a brand attribute (something you declare rather than something you demonstrate), the market for that attribute will follow the same dynamics as every other unregulated market. It will be captured by whoever makes the loudest claim at the lowest cost. The people doing actual, accountable human work, making actual human decisions with actual consequences, do not need a badge. They have receipts. Everyone else is just washing.