AI hiring data lineage is people analytics' next audit test

The record, not the policy, is what gets requested

The Mobley v. Workday matter has kept vendor-versus-employer liability for algorithmic screening in active coverage, with renewed analysis from OutSolve and Legal Examiner following SHRM's earlier framing of the case as a wake-up call for HR. Strip away the procedural detail and a practical question remains for every employer running automated screening: who holds the applicant-level data, and can the employer produce it without the vendor's cooperation?

That question is not answered by a responsible-AI statement. It is answered by a query. And the people who write the query report into people analytics, not legal.

Compliance teams ask for a policy. Plaintiffs, auditors and regulators ask for a record. The record is a data problem, and it lands on people analytics.

Why 2026 guidance points at documentation, not tool bans

Law firm outlooks are converging on posture rather than prohibition. Epstein Becker Green's workplace AI regulation outlook for 2026 and Bricker's employment law report both center on documentation and audit readiness as the employer's defensible position, not on removing tools from the stack. That is a meaningful signal for budget owners: the expected control is evidence, and evidence has a storage location, an owner and a retention clock.

Third-party assurance is maturing alongside it. Warden AI has published findings drawn from more than 150 hiring bias audits, which means audit practice is no longer a thought experiment and auditors arrive with standard data requests. If the employer cannot supply inputs, scores and model versions on request, the audit stalls at the extract stage.

Agents are entering the stack faster than the lineage behind them

HR Tech 2026 is in session this week with an agenda built explicitly around real-world AI implementation, skills and workforce transformation, and vendors are shipping accordingly. UKG's frontline workforce agents were reported by HR Tech Edge during the conference, which is the pattern to watch: agents landing in production operations now, not in a future release cycle.

The business case is easy to justify. HR practitioners report that roughly half of their working week goes to repetitive work, per HR Dive's weekly numbers roundup, and screening and intake are the obvious candidates for automation. S&P Global Market Intelligence's HR tech forecast names people analytics and talent intelligence among the sector's growth drivers, which means more systems generating decision data every quarter. Josh Bersin has argued in the past month that the remit of data and people analytics inside HR is shifting. Custody of decision records is a large part of that shift.

Three moves that turn exposure into routine

Start with a one-page lineage map for each scoring system in the stack. Four fields are enough to begin: the input source feeding the model, the score field and where it is written, the storage owner by name, and the retention clock that applies. Most teams discover during this exercise that at least one system has no named storage owner on the employer side at all.

Then run a thirty-day retrievability drill on a single closed requisition. Pick one that is already decided, request the full scoring trail, and time how long it takes to assemble. The output is not a report, it is a number: days to produce. Finally, carry the finding into procurement. Add an export-on-demand clause and a data-return obligation to the next renewal, before the vendor has leverage and after you know exactly what is missing.

Teams that can answer in days rather than months have not eliminated legal exposure. They have converted it into an operational routine, which is the only version of this problem that scales across a growing portfolio of scoring systems.