Responsible AI — Accountability & Human Oversight

A Manager Uses an AI Tool to Draft All 12 Employee Performance Reviews. The Reviews Sound Plausible and On-Point — but Contain Specific Behavioral Examples the Manager Never Actually Observed. What Now?

A real responsible AI and HR compliance scenario — with three decision options and the right answer.

The Situation

A team manager is facing end-of-year performance review deadlines for 12 direct reports. Time is short, and the manager decides to use an AI writing assistant to draft the reviews — providing each employee’s name, role, and a few bullet points of general impressions. The AI generates detailed, professionally written reviews complete with specific behavioral examples, project contributions, and development suggestions.

The reviews are submitted without a detailed review. Two weeks later, one employee — who received a below-average rating — requests a meeting to discuss their review. They specifically ask about a behavioral example cited in the review. The manager has no memory of the incident described. After checking their notes, they confirm it never happened — the AI generated it.

Choice AStand by the review. The overall assessment is accurate even if the specific example was wrong. Correcting the review now will raise more questions than it answers and could expose the manager to further scrutiny.

Choice BDisclose the situation to HR immediately, correct all 12 reviews to remove AI-generated content that cannot be verified, and accept the consequences. The reviews as submitted are not defensible if challenged.

Choice CCorrect only the one review being questioned and quietly revise the others to remove unverifiable examples before anyone else asks.

The Right Call

Choice B — Disclose to HR and correct all 12 reviews.

Choice A means standing behind fabricated evidence as a basis for employment decisions — the most dangerous option legally. Choice C is a partial fix that doesn’t address the systemic problem and creates a record of selective correction that looks worse than the original disclosure. Choice B is the most difficult option in the moment and the only defensible one. Performance reviews tied to compensation, promotion, or termination decisions are legal documents. Fabricated content in a submitted review creates exposure under employment discrimination law, wrongful termination, and breach of duty of care — regardless of whether the overall rating was accurate.

AI hallucination in performance contexts is particularly dangerous because it sounds authoritative.

AI tools generate plausible content by predicting what sounds right given the context — not by retrieving verified facts. In performance review contexts, this means AI will generate specific-sounding behavioral examples, project outcomes, and interpersonal observations that are internally consistent and professionally phrased but may have no basis in reality. The more detailed and specific the AI output sounds, the more authoritative it appears — and the less likely a time-pressured manager is to question it.

The manager, not the AI, is accountable for what was submitted.

“The AI wrote it” is not a defense in an employment dispute. The manager’s name is on the review. The manager submitted it. The manager is accountable for the accuracy. This is the core accountability principle of responsible AI use in any high-stakes decision context — AI can assist in drafting, but the human who submits the output takes ownership of its contents.

Performance reviews affect employment decisions with real legal consequences.

A below-average performance rating supported by a fabricated behavioral example can affect compensation, promotion eligibility, and, in some cases, termination decisions. If that employee is a member of a protected class and the fabricated example is used as documentation in a subsequent adverse employment action, the organization faces significant discrimination liability. The manager’s time pressure does not reduce the legal significance of what was submitted.

Can managers use AI tools to assist with performance review writing?

Yes — with appropriate oversight. AI can legitimately assist with structuring language, improving clarity, and drafting based on notes the manager provides. The critical requirement is that the manager reviews each specific claim, behavioral example, and factual assertion before submission and verifies that they are accurate and documented. AI-assisted drafting, with human verification, is appropriate. AI-generated content submitted without substantive review is not.

What is AI hallucination, and why is it a risk in HR decisions?

AI hallucination refers to the tendency of large language models to generate content that sounds plausible but is factually inaccurate or fabricated. In HR contexts, hallucinated content is especially dangerous because it appears in documents that carry legal weight — performance reviews, disciplinary records, and termination documentation. An employee subjected to an adverse employment decision based on a fabricated behavioral example has a legal claim regardless of the tool that generated the content.

What should an organization’s AI acceptable use policy say about performance reviews?

Best practice is to explicitly address high-stakes decision contexts — performance reviews, disciplinary actions, promotion decisions — in the AI acceptable use policy. The policy should require that any AI-assisted content in these documents be reviewed and verified by the responsible manager before submission, that specific behavioral examples must be documented in the manager’s own records before inclusion, and that the manager’s submission of the review constitutes attestation of its accuracy.

How to Use This Scenario in Training

Recommended for all managers and HR business partners. Particularly important for organizations where AI writing tools are widely used and performance review periods create time pressure. The key recognition skill is understanding the accountability principle — the person who submits the document owns it — and the specific risk of AI hallucination in behavioral assessment contexts.

This scenario demonstrates a core principle of the Decision Readiness Engine™: the rationalization (“the AI wrote it, the output looks accurate”) is what makes the wrong call feel reasonable. Naming that rationalization — and training the pause before submission — is what builds the accountability behavior this scenario is designed to reinforce.

More Responsible AI Scenarios

Want AI Accountability Scenarios in Your Program?

Xcelus builds responsible AI training that covers accountability, hallucination risk, and human oversight for high-stakes HR and business decisions.

View the Compliance Reinforcement Kit →
Contact Xcelus