133 Job Postings, 110 Internal Notes, 102 Candidates. Also Known As: What Happens When We Don't Understand AI
Let me tell you a story about being a name on a list.
Because for 16 months, I was one. Somewhere in 743 applications, I was a row in a spreadsheet, a match-probability score, a resume getting parsed by a machine that decided in half a second whether I cleared the bar. I know what it is to be reduced to a data point in some company's hiring process. So many of us do by now. You send yourself into the void and you trust (because what other choice is there) that the people on the other end are handling you and your information with some basic care.
So when a LinkedIn connection sent me a handful of links awhile back, I felt my stomach drop. Because what was sent to me, what I found, wasn't a hiring process handling people with care. It was a hiring process that had turned itself inside out and left everyone's private information hanging in the wind.
What I Found
A recruiting firm based out of Ontario: David Joseph & Company. HR consulting, executive search, the whole workforce-solutions pitch. They run searches and sourcing for dozens of early-stage tech startups, and they post those jobs on a standard hosted careers page. Nothing unusual so far.
Except the job descriptions weren't exactly job descriptions.
I went through all 133 of their live postings.
110 of them weren't public job ads at all: they were the firm's internal recruiting file, posted onto the open internet for anyone with a link to read. No password. No login. Just ... there.
And here's what "there" means:
Candid, unfiltered "green flag/red flag assessments of candidates." The blunt internal shorthand recruiters use about people, never for them. Language like "Currently in-between roles. Hard exclude." and "Preference for work-life balance ... hard mismatch with culture."
"Do Not Source" lists naming specific competitor companies. In at least one case, the firm was publicly telling one client not to poach from another client listed on the very same board.
Internal intake notes, comp bands, equity, and (my personal favorite) culture tells like one company's "beyond 9-9-6, founder lives above the office."
And the one that really made me sick: on 34 of those postings, the real full names of 102 individuals the firm had targeted as candidates. Listed. By name. With their current employers and a stranger's one-line judgment attached. Under a heading that literally said DO NOT CONTACT.
One hundred and two people who never applied. Never consented. Never had the faintest idea that their name was sitting on a public page next to someone's assessment of whether they're worth a phone call. The "Do Not Contact" label is pretty telling itself; that section was never meant for daylight. It was inside the machine, but the door to the machine was wide open.
Let me be clear: I'm not printing their names here, and I'm not printing the startups' names either. Those people didn't do a single thing wrong, and neither did the companies who trusted this firm. They're the ones this unnecessarily landed on, and splashing their names across the internet a second time just to make my point would make me exactly the thing I'm writing about. (The difference between exposing a problem and becoming one is whether you're willing to protect the people caught in it. I'll die on that hill.)
But Jillian, Surely This Was Just a Glitch?
Sure. In the same way a wide-open bank vault is "just" an unlocked door. Let's talk about how it actually happened, because this is the part that matters far beyond one sloppy firm.
So, here's the technology stack in play.
David Joseph & Co. uses an AI recruiting platform called Contrario; YC-backed, the whole "AI agents plus expert recruiters" pitch. You do an intake call, you upload your notes, and Contrario's AI builds a rich internal brief for each role: match scores, calibration, do-not-source lists, and the named prospect list. That brief is the recruiter's digital brain. You can literally see the AI's fingerprints on these posts: little machine-generated hedges like "for reference only–do not source these specific profiles" that belong in a back office, not a job ad.
Then there's a separate system: Manatal, a hosted careers-page product, that publishes the actual public listing candidates see and seemingly acts as an ATS.
The failure happened in the two inches of coded space between those two systems. Somebody (or more likely, some misconfigured automated sync) took Contrario's internal brief and shoved it straight into Manatal's public description field. The AI did it's job perfectly. It built exactly the brief it was supposed to build. The catastrophe is that nobody drew a line between the machine's private notes and the page the world sees. And additionally: no one bothered CHECKING the job postings once they went live. Technology glitch, meet human incompetence.
And this is the whole ballgame now in 2026, so I'm going to state it plainly:
The danger of AI at work is almost never the AI. It's the shoddy plumbing you bolt it to.
A human being carelessly mishandling one job post exposes one job post. A bad integration mishandling the template exposes the entire board: 110 roles, 66 clients, 102 human beings in a single stroke, and then does it again every time a new role gets posted. The carelessness came from human error, automation merely accelerated it. Somewhere, there was a "Pending Approval" gate sitting right there in the workflow, and it approved all of it, because nobody built it to ask the only question that actually mattered: "would I be okay if the person we named read this?"
Why This Gets Under My Skin
I write a lot about how business-level decisions land on individual people. It's kind of my whole thing. A boardroom decides on a headcount reduction and eight months later somebody they never actually met is filing Chapter 7. The decision is abstract; the cost is a specific human being with a specific mortgage.
This is that, but somehow even more intimate. Because recruiting, REAL recruiting, is a trust business. Strip away the software and the fees and the AI agents, and the entire job is this: people hand you sensitive information about themselves, often at the most vulnerable moments of their working lives, and you are supposed to be a careful steward of it.
That's it. That's the job. Everything else is corporate performance.
And when a firm collects the fees, bolts on the shiny AI tool, and can't even be bothered to check whether it's broadcasting a hundred people's names to the internet; that's not a technical oopsie. That's a firm that decided the "careful steward" part was optional. That treated the human beings flowing through its pipeline as data exhaust instead of people. And it's a preview of a thing that's about to happen everywhere, in every industry racing to duct-tape AI onto processes they never bothered to understand.
People-First Isn't Just A Tagline
I'm not anti-AI. I want to be clear about that. I'm anti- using AI to mask your own carelessness and negligence. If you're going to run a people business (and recruiting is one of the most people businesses there is) here's the lowest bar to clear:
A person's data belongs to the person. Their name, their employer, the fact that they're privately open to a move–that's THEIRS. Holding it is a privilege with obligations attached, not merely a byproduct of running a search.
Build like the internal will become external, because one bad field-mapping is all it takes. If a note would be indefensible on a public page, the fix isn't to hide it better. It's to actually ask why you're writing it that way at all.
Own the whole chain. You don't get to hide behind "the tool did it." If your AI sourcing platform feeds your public career page, then vetting the wire between is the job. It's the work at-hand, for a human to execute.
And when you get it wrong (and everyone eventually gets something wrong) you move fast for the people you exposed, not for your own reputation. You pull it down. You warn the people caught in it. Then you worry about how you look.
Let's Bring It Home
Standing up a firm and collecting checks is the easiest part. Anybody can print business cards and buy a software subscription. The hard part, the part that actually EARNS the fee, is being the kind of steward that a guarded, mid-search human being can hand their information to without flinching.
A firm that treats that as an afterthought is running a lot more than a discounted version of recruiting. It's not running recruiting at all. It's collecting payment while letting private, sensitive data get spilled with every "post job" click your recruiters do.
There are 102 people out there right now whose names sat on a public page because a company was more excited about the AI than it was careful about the humans. Most of them still don't know. Some of them are getting a very strange message from me.
And the rest of us should take the warning for what it is: the AI isn't going to be the thing that hurts people. The people who bolt it on and walk away; they're the thing. They always have been.
So here's the question I'll leave you with. When the tool makes it this easy to forget that the name in the system belongs to an actual person, are you going to be the one who remembers anyway?
The specifics here are drawn from postings that were publicly accessible at the time of review. Individuals and client companies have been deliberately left unnamed. The firm responsible and the platforms involved have been notified.
NOTE: At the time of publication, Contrario has corrected the issue and the public job postings no longer contain the internal notes.
When reached for comment, Contrario said: "We investigated the specific issue you referenced and confirmed that the affected content has since been removed from the public page. Based on our review, this does not appear to reflect a broader Contrario platform issue, though we're continuing to review the relevant workflow out of an abundance of caution."
Both David Joseph and Manatal were reached for comment and did not respond.
Comments