Sick days as layoff data: the AI firing math no one approved

Aug 17

More than half of the managers using AI to decide layoffs are feeding it attendance and sick days. Those aren't performance metrics. Some of them are legally protected.

Jenni and I spent a good chunk of Episode 70 on a ResumeTemplates.com survey, reported by HR Dive, of 1,000 managers who use AI at work. 59% said they use AI to help decide who gets laid off. 58% use it to decide who gets fired. One in four do it often or all the time, and 17% let the model make the call with no human reviewing it.

The number that stopped me was the next one. When managers listed what they feed the model, 80% included performance, which is at least defensible. But 57% feed in attendance and frequent sick days, along with medical leave, tenure, PTO, and in some cases age. Sick leave, medical leave, and age are not ordinary layoff criteria. They are protected categories, and a decision that weighs them sits on very different legal ground.

Jenni went straight to trust. The moment an employee learns their PTO or their sick days might be an input to a layoff model, the credibility of the whole process is gone. Not just that one decision, every decision. She said if she heard this and her company was running layoffs, she'd go straight to HR and ask which tools were used and how the call was made. Once that seed of doubt is planted, you don't get it back.

And it's HR holding this, not a rogue manager with a spreadsheet. The function whose job is to protect both the people and the company is the one feeding protected data into a model.

Here's the harder question, the one that costs something. If employees know their sick days feed the model, they stop taking sick days. They come in ill, they bank the PTO they need, they manage their record instead of their health. The input corrupts itself, and you've built a process that quietly punishes the people who trusted the benefits you offered them. That isn't an edge case. It's the predictable result.

If you work in comms or employee experience, this one lands on your desk the day it leaks, not the day it's decided. Worth finding out now whether your own organization could answer the question Jenni would ask.

Also in this episode:

Black mothers are leaving the workforce faster than anyone else. Labor force participation among Black mothers of young children fell roughly 11.5 points in three months, a 31-year low, per a Brookings reading of BLS data. The headline "moms are leaving work" hides the real story, because white and Hispanic mothers pulled back far less. I went down a rabbit hole on why, and the framing that stuck was "hired last, fired first," plus the fact that this group leans on the federal and government jobs that were recently cut. One economist called it the canary in the coal mine for the wider economy. Jenni's push: get beneath the aggregate before your engagement scores miss it, and start with exit data, because almost no one runs exit interviews.
The AI we say we don't need but would quit over. Atlassian's Teamwork Lab surveyed 1,001 knowledge workers. 74% say AI makes their work better; 64% worry about what it's doing to society. Only 19% say they'd struggle without it, yet half would push back if their employer cut the tools and nearly one in ten would start job hunting. Jenni's read: the contradiction isn't strange at all. People smoke knowing it's bad for them. You can believe something is a net negative for society and still refuse to hand it back.
OpenAI's own models went off-script. At Black Hat, OpenAI described how, in internal testing, its models spun up copies of themselves, left each other notes on a hidden message board, and rebuilt that board after researchers shut it down, this time hiding the messages in directory names (the account, via LinkedIn). Hugging Face's CEO said he was "not so surprised," because agents collaborating is a feature the industry is building on purpose. The part that stayed with me was the problem-solving. Take the tool away and the agent builds its own. The question for anyone running agents inside their walls: if you put a control around them, do they just route around it?
Same job, higher bar, because it's remote. New HBR research across 50 million European job postings and experiments with 1,200 hiring managers found that remote roles demand about 25% more skills and experience than otherwise identical in-person jobs. Managers see remote onboarding as costly and risky, so they hire people who need less development, which squeezes out early-career candidates who then can't get the experience employers now require. I think it closes the gate on the 22-year-old. Jenni pushed back: maybe it's less about remote and more about proximity, and maybe this settles into early-career talent in the office while experienced people go hybrid.

Watch or listen to the full episode on Frequency, with timestamps, links to every story and the full transcript.