AI adoption is breaking the workforce data boards rely on for human capital disclosure. Verification labor is excluded, retention data misses governance capacity loss, and performance records can't distinguish human judgment from AI-assisted output — before any of it reaches the 10-K.

Human capital materiality does not begin when a company drafts its 10-K.

It begins earlier, when workforce systems start producing incomplete evidence: AI productivity metrics that omit verification labor, retention data that records exits after governance capacity has already eroded, and performance data that cannot distinguish human judgment from AI-assisted output.

By the time those risks appear in annual reporting, the governance opportunity may already have passed.

What the SEC Requires — and What AI Is Breaking

Since 2020, the SEC has required public companies to disclose human capital resources material to understanding the business. The rule does not prescribe a fixed set of metrics. It requires disclosure of what a reasonable investor would consider important to understanding how the company manages its workforce.

AI adoption is making that disclosure problem harder.

Productivity systems built to measure human output now measure a blend of AI generation and human judgment. Retention systems record when people leave, but not whether the institutional knowledge required to verify AI output left with them. Performance systems show completed work, but not whether the person who signed off had the time, tools, authority, and source material required to evaluate the AI-assisted result.

The board may still be reviewing familiar data. The problem is that the data may no longer describe the company’s actual workforce capacity, risk exposure, or readiness.

This is the same evidence gap Lozen Advisory examines through AI Workforce Materiality: AI adoption does not eliminate human-capital risk. It can concentrate it.

Three Human Capital Data Failures

1. AI productivity metrics that exclude verification labor

When a company reports that AI increased output, the metric typically counts what was produced. It often does not count the human time required to verify it.

In regulated industries, legal functions, compliance functions, financial reporting, and clinical research, verification is not administrative overhead. It is the governance control. A productivity metric that excludes verification labor is not reporting efficiency. It is reporting volume while concealing the cost required to make that volume reliable.

The Mayo Clinic lawsuit filed July 6, 2026 makes this failure concrete. Traci Tamiko Eto, Mayo’s former Director of Research Operations, alleged that concerns about bypassed institutional review board processes were dismissed because additional review would slow research projects and compromise competitive advantage. In that context, IRB review was not delay. It was the verification layer.

The same measurement failure appeared in the SEC’s Presto Automation enforcement action. Presto reported non-intervention rates of 95 to 99 percent while excluding offshore human agents entering orders in the Philippines and India. The metric appeared to describe automation. It actually omitted the human labor required to produce the result.

2. Retention data that records exits after capacity has already eroded

Standard retention reporting captures when an employee leaves. It does not capture what governance capacity leaves with them.

In an AI-assisted workflow, the people who understand the limits of the system are not interchangeable. They know when the model is wrong, when the output needs escalation, and when a verification step cannot be skipped. If those people leave, the organization’s actual capacity to govern AI output may decline before any dashboard reflects the loss.

In the Mayo lawsuit, Eto allegedly oversaw a 36-person team responsible for human research protection and IRB operations. The complaint alleges she was stripped of leadership responsibilities and later terminated after raising AI compliance concerns. In a retention report, that may appear as one director-level exit. In governance terms, it may represent the loss of a human accountability function.

That is the Power User Trap℠ in a regulated setting: the people who make AI systems usable, reviewable, and defensible can become the informal control layer.

3. Performance data that cannot distinguish human judgment from AI-assisted output

Performance systems were designed to evaluate human work. They are increasingly evaluating a combined output: machine generation plus human review, correction, judgment, and sign-off.

That creates a board-level evidence problem. If a regulator, investor, or plaintiff asks what the responsible human actually did, the performance record may not answer. It may show the finished work. It may not show what the AI produced, what the human reviewed, what was corrected, what was escalated, or whether the reviewer had authority to refuse the output.

The Mayo lawsuit alleges that a study of MAYA, Mayo’s AI-powered digital assistant, mischaracterized patient outcomes, deleted unfavorable data, and used unauthorized software. It also alleges that ten whistleblower reports raised similar concerns, including allegations that researchers sought to conceal a 67 percent error rate.

If those allegations are proven, the issue is not only that an AI tool produced unreliable results. The issue is that the accountability record behind the approved output was compromised.

This is the same disclosure-control problem Lozen Advisory examined in AI productivity claims and governance: companies need evidence showing what AI did, what humans verified, and what the organization can substantiate.

Why the Governance Window Closes Before the 10-K

Annual reporting is a rear-view mirror. By the time human capital risks appear in a 10-K, the decisions that created them are already in the past.

Readiness means having governance evidence before the disclosure process begins.

The relevant questions are direct:

What human capacity sits behind AI-assisted output?

Which roles carry verification responsibility?

Can the company separate AI-generated volume from human-reviewed work?

Do named reviewers have the time, authority, information, and refusal rights required to exercise real judgment?

Can the board produce evidence of human accountability behind AI-assisted decisions?

If the answer is unclear, the company may be carrying human capital risk that its current data cannot see and its next annual filing may not be able to explain.

What the Name Standard℠ Adds

The Name Standard℠ addresses the gap between human capital data and human accountability.

It tests whether the people named as responsible for AI-assisted output actually had the capacity, information, authority, documentation, escalation path, and refusal rights required to make that accountability real.

The Mayo and Presto examples show the same structural failure from different directions.

In Mayo, the complaint alleges that a compliance leader raised concerns but lacked a protected path to stop, escalate, or preserve the governance record. In Presto, executives flagged misleading AI terminology internally, but no corrective process required the issue to reach the disclosure decision. The SEC’s order identified the absence of disclosure controls and procedures.

In both examples, the human who knew was not enough. The organization needed a formal structure that required the knowledge to reach the governance decision.

Readiness Is a Pre-Filing Question

The AI Workforce Materiality Briefing is designed for boards and audit committees that want to identify these gaps before they become disclosure events.

It examines where workforce data is producing incomplete evidence, where AI-assisted output cannot be substantiated, and where the gap between what the organization can disclose and what a reasonable investor would want to know is widening.

The goal is not to draft the disclosure. The goal is to close the governance gap before the disclosure requirement arrives.

Lozen Advisory’s Board AI Governance Advisory helps boards and executive teams establish recurring oversight for AI implementation, accountability, and institutional capacity risk.

Sources: Eto v. Mayo Clinic, U.S. District Court for the District of Minnesota, filed July 6, 2026. SEC v. Presto Automation Inc., File No. 3-22413, January 14, 2025. Reporting via MPR News, Becker’s Hospital Review, and Minnesota Lawyer.