Most ethics failures inside large organizations are not failures of character. They are failures of modeling. An employee who discovers that an AI system's public performance claims do not match its internal test results makes the decision to surface that gap or let it pass in seconds, not months. What determines the choice is rarely the employee's integrity. It is what the employee has watched the organization reward and punish in the eighteen months before that moment.
This is the distinction most ethics governance programs get backward. They spend their budget on the declaration: the code of conduct, the annual training module, the values statement in the onboarding deck. They spend almost nothing on the evidence of what the organization actually does when someone acts on the declaration. An organization can have a flawless code of conduct and a culture that quietly punishes the people who follow it. Both things are true at once, in the same building, often inside the same fiscal quarter.
The Safety Gap Is the Exposure, Not the Training Gap
Every organization I have watched fail an ethics test had already passed the training test. The employees could recite the code of conduct. What they could not do, and what no training module teaches, is predict what happens to the next person who surfaces a gap between what the company claims and what the company's own data shows.
That prediction is built from evidence, not policy. It is built from watching whether the last person who raised a concern was thanked in a way the rest of the team could see, or quietly reassigned in a way the rest of the team could also see. Employees are exceptional pattern-readers. They do not need the organization to say retaliation is acceptable. They only need to observe one unresolved case where surfacing a gap cost someone their standing, and the lesson propagates through the team faster than any policy update.
This is why the financial incentive to bypass the internal channel entirely keeps growing. The SEC's Whistleblower Program, created under Dodd-Frank Section 922, pays awards of 10 to 30 percent of sanctions collected in enforcement actions above $1 million, funded from a dedicated investor protection reserve, with cumulative payouts now exceeding $2 billion. An employee who has watched an internal escalation get rationalized away is not choosing between silence and conscience. They are choosing between an internal channel with a known bad outcome and an external one with a defined financial one. Organizations that have not made the internal channel visibly work have built the exact conditions that make the external channel rational.
Why the Pattern Survives Good Intentions
The pattern persists because declaring a value and modeling a value require different things from leadership, and most leadership structures are built to do the first and incapable of the second.
Declaring a value costs a leadership team an afternoon: draft the language, approve it, publish it. Modeling a value costs something continuous: it requires a leader to accept friction from their own product timeline when a claim has not been verified, to publicly credit the employee who slowed a launch rather than quietly note the delay as a problem, and to apply the same standard to their own team's claims that they apply to everyone else's. Most leadership teams will do the first indefinitely and the second only when it is free.
The 2026 Edelman Trust Barometer found that 70 percent of people refuse to trust an institution whose stated values diverge from its observed behavior. That is not a marketing statistic. It is a description of exactly what an employee is calculating in the seconds before they decide whether to raise a hand. The gap between what leadership says and what leadership visibly does is not a communications problem. It is the entire mechanism by which a culture either produces gap-surfacing or produces rationalization, at scale, without a single written instruction telling anyone to look away.
What Modeling Actually Requires
There is a useful distinction between two objectives an organization can build toward, and naming it changes what leaders build. The compliance-minimum objective produces a culture that avoids getting caught: employees know the policy, understand that violations are taken seriously in theory, and have a channel to use if they choose to. This objective reduces isolated ethical lapses. It does not touch the deeper failure, because it never asks leadership to change its own behavior.
The elevated objective is different in kind. It is the Expectation Elevation Model applied to culture rather than performance: a leader's job is not to manage compliance but to raise what people believe is possible and rewarded inside the organization. Under this objective, the first question in a product review is not “will this get us in trouble,” which is a compliance question answered by lawyers after the fact. It is “have we verified this claim,” which is a competency question answered by the team before publication. Only the second question, asked consistently and visibly by leadership, produces a culture where gap-surfacing is normal rather than heroic.
I have watched the difference play out inside organizations that looked identical on paper: same code of conduct, same training vendor, same values poster in the lobby. The difference was never in the document. It was in whether the senior leader in the room, faced with their own team's unverified claim, treated the delay as the cost of integrity or as a problem to be managed around. Employees do not learn ethics from the poster. They learn it from watching what happens to the people above them when the standard gets expensive.
The organization that builds a culture where gap-surfacing is visibly rewarded, where the named accountable person has real authority rather than advisory standing, and where every leader has personally absorbed the cost of their own verification standard at least once, has built something that holds without anyone reciting the code of conduct. The organization that has only declared the standard has built a document that will read very well in a deposition and explain nothing about why the gap existed in the first place.
Touch Stone Publishers works with boards and executive teams who want to close that gap before a regulator, a plaintiff, or an employee's own account of it does the job for them.
Culture is not declared from the top. It is modeled, or it does not exist.