AI's Do's and Don'ts Are Not Enough. You Also Need an Examination of Conscience.
Usage guidelines tell you what not to do. They rarely tell you what you owe the people affected by what you do
Every organization adopting AI eventually produces a list. Do verify outputs before relying on them. Don't input confidential information into public tools. Do disclose AI assistance where relevant. Don't let the tool make decisions that require human accountability. These lists are useful, and every institution that uses AI at any scale needs one. But a list of do's and don'ts, however well constructed, answers a narrower question than the one that actually matters: not "what are the rules for using this tool," but "who am I becoming, and who is affected, by how I use it."
Rules Are a Floor, Not a Formation
A compliance checklist can prevent the most obvious harms. It can stop confidential data from leaking into a public model, stop an unverified claim from reaching a customer, stop a decision that requires human judgment from being quietly delegated to a system incapable of exercising it. This matters, and no organization should skip it. But a checklist, by its nature, addresses only the failures someone has already anticipated. It cannot form the habit of asking new questions as new situations arise, and AI produces new situations constantly, faster than any list can be updated to cover them.
What the checklist model misses is closer to what the Christian tradition calls an examination of conscience, a habit of periodically asking not "did I break a rule" but "did I treat the people affected by my actions with the attention their dignity deserves." Applied to AI use, this is a different and harder question than any do-or-don't item can capture. Did I let the tool's fluency substitute for my own judgment when the stakes were high enough to deserve better? Did I use a shortcut on a task where the person on the other end needed to know a human being actually attended to their situation? Did convenience quietly become the standard by which I decided how much care something deserved?
The Habits a List Cannot Instill
A rule against inputting confidential data is easy to follow and easy to verify. A habit of pausing to ask whether a task actually warrants full human attention, regardless of what any rule technically permits, has to be formed, not merely posted on a wall. This is the difference between compliance and virtue. Compliance asks whether an action fell inside the permitted boundary. Virtue asks whether the action, boundary aside, was actually good, whether it served the flourishing of the people it touched.
Organizations serious about using AI well need both. The do's and don'ts protect against the failures everyone can already see coming. The examination of conscience protects against the failures nobody wrote down because nobody anticipated them, which, with a technology this new, describes most of the failures still ahead.
A Practice Worth Adopting
Alongside any usage policy, a simple discipline is worth building into how teams work with AI: a regular, honest pause to ask not whether the rules were followed but whether the people on the receiving end of the work were actually served by how it was done. This cannot be automated, delegated to the tool, or satisfied by a checkbox. It requires the same thing every genuine moral discipline requires, a person willing to ask the harder question of themselves before anyone else asks it of them.
The list of do's and don'ts will keep growing as the technology changes. The examination of conscience underneath it does not need to change at all. It only needs to be practiced.