Meta's RIF model just became evidence

Most platform evaluations start with a demo request and end three weeks later in a conference room. This one takes 15 minutes and puts you directly inside Gladly's interface — navigating it on your own terms.

See how AI surfaces real-time customer context before a conversation starts. Watch how a single conversation thread pulls in purchase history, channel history, and account details without a handoff.

No installation. No commitment. Start the interactive demo and see the platform for yourself.

See it yourself

On July 15, 2026, twenty-six former Meta employees filed a complaint alleging that Mark Zuckerberg's 8,000-person restructure used an AI scoring system that disproportionately flagged workers on medical, parental, and caregiver leave. The complaint names the model. It names the vendors. It names the HR leaders who signed off on the output without an audit trail. The filing is still moving through the news cycle this week.

The specific claim is not that the algorithm was designed to discriminate. It is that no one at Meta ran the disparate-impact test before the layoff list went to legal. Workers on protected leave were statistically over-represented on the "low performance signal" tier — and the company cannot show a document proving anyone checked.

Every algorithmic HR decision your team made in 2026 is now discoverable, and the burden just shifted from the plaintiff to you. This is not a compliance story — it is the moment the AI Wage Gap becomes a legal-liability gap. On one side sit the organizations that can produce an audit log, a bias test, and a human sign-off for every model output that touched a person's job. On the other side sit the organizations that trusted the dashboard and shredded the workpapers.

The CHROs who ran headcount planning through a scoring model this year now have a homework problem their CFOs did not price into the restructure. It is not "did we discriminate." It is "can we prove we checked." Those are very different documents, and only one of them exists on most enterprise laptops right now.

This week, audit ONE algorithmic HR output — the last scoring run your team used for RIF, promotion, comp, or performance flags — against protected-class distribution before Friday. Pull the output. Pull the pre-model population. Compute the ratio for each protected class. Write down the delta. If the delta is greater than the 4/5ths rule threshold and there is no human sign-off memo attached to the run, you have a Meta problem, not a Meta headline. The artifact is a single one-page memo: model name, run date, disparate-impact ratio, human reviewer, decision. That memo is the difference between a tool and an exhibit.

The operators already doing this stopped calling it "AI governance" months ago. They call it defensible decisions — and their scoring models are still running, while everyone else's are about to get subpoenaed.

You just found the one model that needs a memo. See how many others are hiding in your stack → Take the BEAST Score

📊 Get your BEAST Score — the free 0–100 read on your AI Wage Gap. Take the BEAST Score →

🎯 Portfolio Executive Cohort — limited seats.Build the artifact, not the policy →

The Leverage Signal is written by Yuri Kruman: 3x CHRO/CLO, AI trainer at Meta, Microsoft and OpenAI, founder of Portfolio Leverage Co. and author of the forthcoming Closing the AI Wage Gap. Reply to this email. I read every one.

📬 Forwarded this? Subscribe free → One signal, one move, every weekday.

🧭 Work with Yuri