The Dark Side of Mercor You Should Know Before You Start

I've worked on 10+ Mercor projects. Here are the five patterns nobody warns you about — and the preparation system that keeps me getting paid. Inside the Data Factory, Episode 1 of 4.

In November 2025, Mercor’s twenty-something founders became, by Forbes’ count, the world’s youngest self-made billionaires, after raising $350 million at a $10 billion valuation. About a week later, thousands of contractors on one of its largest AI training projects woke up locked out of Slack. The project had been cancelled by email. Some were offered their jobs back, at roughly a third less per hour.

I’m not writing this from the outside. I’ve onboarded onto more than ten Mercor projects: rating model responses for a frontier lab, writing evaluation rubrics for consumer tasks, building agentic coding challenges, recording my own screen while completing real-world errands so a model could learn from them. I’ve been through the welcome calls, the unpaid assessments, the quiz-gated onboardings, the silent offboardings. I’ve read more “you’ve been carefully selected” emails than I can count.

Here’s the strange part: it’s still some of the best-paying flexible work available to a knowledge worker right now.Real rates, weekly payouts, work you can do from anywhere. That’s exactly why the dark side matters. The money is real, and so are the traps, and nobody at Mercor is going to explain the traps to you, because the system works fine for them whether you figure it out or not.

This is Episode 1 of a four-part series. Today: how to recognize the five patterns that burn new contractors, and, for paid subscribers, the preparation playbook I wish I’d had before my first project. Episodes 2 through 4 go deeper: the AHT metrics game, the cat-and-mouse of audits and enforcement, and how to stack projects into something resembling stable income.

Let’s start with what you’re actually walking into.

Part 1: The five patterns

1. You are "carefully selected," and instantly disposable

Every project welcome doc opens the same way: you’ve been carefully selected, you’re joining something first-of-its-kind, the team is so excited to have you. Take the flattery at face value and you’ll misread your position entirely.

The operating unit at Mercor is the project, not the person. Projects appear with 48-hour pilot deadlines, pause without explanation, resume in “waves,” and vanish mid-sprint. Onboarding is random: you might get a live call, or just a 40-minute document and a quiz. Offboarding is more random still: often no email at all, just a task queue that never refills and a Slack channel that goes quiet. The Forbes-reported mass cancellation wasn’t an aberration; it was the normal lifecycle, at unusual scale. When a frontier lab’s priorities shift, your project shifts with it, and you’re the shock absorber.

The mistake new contractors make is treating a project like a job. It’s not. It’s a gig with a countdown timer you can’t see.

2. The clock decides everything, and the clock is ambiguous

Mercor projects run on billed hours and throughput metrics, most importantly AHT, the average handling time per task. Here’s the trap: the rules about what counts as billable are strict, but the definitions are fuzzy, and the enforcement is asymmetric.

One project guideline I worked under stated flatly that charging time outside of onboarding and active task work would result in contract termination and would affect future work with Mercor. Fair enough. But what is active work? The guideline doc itself takes 30-40 minutes to read. That reading is paid once, though the doc changes weekly. Re-reading the changelog? Gray area. Waiting for a task to load in a laggy annotation tool? Gray area. Asking a clarifying question in Slack and waiting 20 minutes for an answer? Gray.

Meanwhile the asymmetry: bill too many hours relative to your output and you’re flagged for overcharging. Bill too few and your AHT looks suspiciously fast, which reads as low-effort work. There is a corridor of acceptable speed, nobody tells you where its walls are, and people get terminated for hitting either one. Episode 2 is entirely about finding that corridor.

3. Slack is the real performance review

Every project has a Slack workspace, and every guideline doc has a section on Slack etiquette: where to post, how fast to respond, what not to ask. Read those sections as what they really are: a warning that you are being evaluated in Slack at all times.

Ops leads are drowning: hundreds of contractors, client pressure, shifting specs. The contractors who survive project transitions are the ones ops can see: present at announcements, asking one sharp question instead of five lazy ones (nothing marks you faster than asking something answered in the doc), reporting genuine tool bugs clearly. Slack is also your early-warning system, where pauses, wave assignments, and quality crackdowns show up hours or days before any email. Contractors who treat Slack as optional get offboarded confused. Contractors who treat it as a stage get invited to the next project.

4. The rules change underneath you

Guideline docs are living documents, and I mean that as a threat. Changelogs, rubric refreshers, v1.1 becoming v2.0 mid-sprint, “golden examples” added in week three that quietly contradict what week one told you. The work you submitted Monday can be reviewed Thursday against standards published Wednesday.

This isn’t malice. It’s the client lab refining what it wants in real time, with Mercor passing every revision straight through to you. But the effect is the same: your quality score, your review outcomes, and therefore your continued employment are graded against a moving target. The contractors who get burned are the ones who read the guidelines once, at onboarding, and assume they’re done. The document is never done.

5. The cat-and-mouse game

Layered over all of it is mutual surveillance. Mercor and its clients police contractors: LLM-usage policies (on projects that are themselves training LLMs, ironically), screen recording requirements, monitoring software, daily audits, quality matrices, error-lookup tools, and reviewer pipelines where another contractor’s judgment gates your pay. Some contractors, in turn, probe every boundary: outsourcing tasks, botting assessments, running AI through AI-detection. The company adapts, the workers adapt, and the honest majority in the middle absorbs the friction: more invasive checks, more hoops, more sudden rule changes aimed at someone else’s exploit.

And the house holds all the data. When Mercor was breached in March 2026, what leaked from 40,000+ contractor files told us exactly what they held: government IDs, banking details, biometric video interviews, and screenshots captured from workers’ devices by monitoring software. That breach triggered a wave of lawsuits, and it should permanently change how much of yourself you hand over. (More on data self-defense below the paywall.)

The macro picture matches the micro: the wage cuts Forbes documented, a trade-secrets lawsuit from Scale AI, and Forbes reporting on fraud rings and North Korean operatives infiltrating the contractor pool. That history is precisely why legitimate contractors like you face ever more aggressive verification.

So why am I still here?

Because the equation still clears. I’ve been paid on time, every week, for every approved hour across ten-plus projects. The hourly rates beat most freelance alternatives. And frontier-lab data work is a genuinely fascinating window into where AI is actually going: you see what the labs care about six months before it ships.

The contractors who thrive aren’t luckier. They walk in with a system: they read the right sections of the guidelines first, log defensively from day zero, manage their Slack presence deliberately, and never let one project become their whole income. The rest of this episode is that system, the part I charge for, because it’s the part that took me ten projects and a few painful offboardings to learn.