Is Your HR Data Authentic? Combating Disinformation and AI Manipulation in People Analytics

HR has become a data discipline almost overnight. Promotions tied to performance scores. Compensation tied to calibration curves. Retention plans tied to engagement signals. The problem: nobody has audited whether the data feeding those decisions is authentic, complete, and resistant to manipulation.

In 2026, that's not a theoretical risk. It's a live one.

Where HR data integrity breaks today

• Engagement scores skewed by coordinated team responses or bot-driven survey clicks.
• Performance reviews drafted by LLMs and lightly edited — homogenized, hard to calibrate.
• Identity fraud at hire — deepfake video interviews and AI-generated work samples.
• Manipulated 360 feedback through coordinated retaliation or anonymous abuse.
• Vendor dashboards using black-box algorithms that change underneath you without notice.

A digital-trust framework for HR data

1. Provenance

Every data point that drives a people decision must carry its source, its collection method, and its last-modified timestamp. If you can't trace where a number came from, it does not enter a promotion conversation.

2. Authentication

Identity verification at hire is now infrastructure, not paranoia. Liveness checks, document verification, and reference-call verification for senior roles. For internal data, single sign-on and audited access logs at minimum.

3. Anomaly detection

Bot detection on engagement surveys. Stylometric checks on bulk-submitted reviews. Outlier detection on calibration shifts. None of this is optional in 2026 — vendors should provide it; if yours doesn't, ask why.

4. Disclosure

Require AI-assistance disclosure on any document that influences a people decision — performance reviews, peer feedback, hiring scorecards. Not to ban it, but to weight it appropriately.

5. Human-in-the-loop on consequence

Any decision with material consequence for a person — termination, denial of promotion, performance improvement plan — requires a documented human review of the underlying data, not just the score.