You Required the AI. How Will You Judge the Worker?

A tech recruiting company’s CEO banned AI-generated internal communications after receiving a flood of AI-written Slack messages, according to Business Insider reporting cited by Futurism. Futurism also cites an anonymous CEO’s threat, reported by Inc, to fire the next employee who sent unedited ChatGPT material. These are specific restrictions on workplace messages, not evidence that every employer has abandoned AI.

The harder problem concerns evaluation. Futurism describes experiments led by Samuel Meyers, a Hebrew University of Jerusalem doctoral student, and published in The Leadership Quarterly. Supervisors decided whether to trust an employee’s report with information about whether the employee had used AI. Futurism reports trust rates of 68 percent without AI use and 24 percent with it. The account says those judgments did not rest on the work’s quality.

Those figures describe experimental judgments, not measured output quality or a count of promotions denied. Futurism reports that the trust penalty also appeared where workplaces encouraged AI. Encouragement, a binding mandate, and permission remain different policies. But the finding exposes a problem any mandatory rollout must address: the organization can reward adoption in its instructions while penalizing adoption in its assessments.

Managers have a legitimate reason to reject messages that waste colleagues’ time. A sender still needs to check facts, remove irrelevant material, and make the request understandable. A ban on unedited output can serve that purpose. It does not establish that every assisted report deserves less trust. The relevant failure is the employee’s failure to review the message, not the mere presence of a tool.

The familiar image of competence complicates the rule. An unaided report may suggest that its author knows the subject and can defend the conclusions. Required assistance changes what that signal tells a supervisor. Employers must test understanding directly instead of demanding modern production and an antique certificate of effort. The software license should not come with a character deduction.

Employers can separate three questions: Did the employee follow the tool policy, did the work meet its requirements, and could the employee explain and verify the result? Evaluation criteria should specify accuracy, relevance, and appropriate handling of sensitive information. Managers should judge those requirements against evidence in the work. AI disclosure should help reviewers inspect the process, not substitute for inspection.

Before requiring adoption, management can publish permitted uses, review duties, and performance criteria. Employers can also explain how workers may challenge an evaluation that treats authorized assistance as misconduct or incompetence. A second reviewer should examine the disputed work against the stated criteria. The next decision belongs to management: define successful compliance before attaching consequences to it.