What Outlier's TOS Actually Says About Account Sharing — And Why It Gets People Permabanned in 2026

TL;DR: Outlier's TOS prohibits sharing your account in any form — credentials, devices, IP overlap, and "assisted" task completion all count. The platform detects shared accounts through device fingerprinting, IP clustering, behavioral biometrics, and quality-signal drift. Bans are issued without warning and are functionally irreversible. If you've been doing any of the things in the second half of this article, stop today. If you've already been banned for it, the appeal path is closed — your next move is parallel applications to DataAnnotation, Alignerr, Mercor, and Telus AI. None of them share contractor records with Outlier.

I want to start with the version of this story that I hear most often, because it's the version that gets people the angriest.

A woman emails me. Her boyfriend got onto Outlier in late 2025. He's a software engineer, picky about his time, never really did the work consistently. She has a master's in linguistics, was between jobs, and started "helping him" — which quickly turned into her doing 100% of the tasks under his login, on his laptop, while he worked his actual job. She was good at it. She made them roughly $1,800 a month for four months.

Then one morning the account closed. No warning. No email subject line that politely suggested anything. Just Account Status: Removed. Reason: Not provided.

She wanted to know if she could appeal. The answer is no. She wanted to know what they could have done differently. The answer is, on a technical level, almost nothing — they did the one thing that gets flagged the fastest. She wanted to know if she could just sign up under her own name now. The answer is also no, and the reason why is what this entire piece is about.

So let's talk about what Outlier's terms of service actually say, what the detection system actually catches, and the specific behaviors that get people removed permanently — including a few that almost nobody realizes are violations.

What the TOS actually says, in plain English

If you have ever read a platform's terms of service in full, you are an unusual person. If you have read Outlier's, you are an even more unusual person, because Outlier buries the relevant clauses under three layers of contractor-agreement boilerplate.

The TOS effectively says four things about your account:

1. Your account is personal and non-transferable. You may not allow anyone else to access it, log into it, or perform work under it. "Anyone else" includes spouses, partners, family members, roommates, and coworkers. There is no exception for "they're just helping."

2. You must complete the work yourself. All tasks submitted under your account must reflect your own work product, on your own equipment, under your own login session. Outsourcing, ghostwriting, AI-assist beyond approved internal tools, or having a friend "polish" your responses are all violations.

3. You may have only one account. This is the duplicate-account rule, and it's enforced by device fingerprint, IP, payment method, tax ID, browser entropy, and sometimes by behavioral biometrics like typing cadence. Even if your roommate creates their own account on their own laptop, if the household IP and payment infrastructure overlap heavily, both accounts can get flagged.

4. You must not misrepresent your identity, location, or eligibility. This is the catchall. It is what they use to remove anyone who used a VPN to apply from a non-supported country, anyone who lied about their education, and anyone whose tax forms didn't match the name on the account.

Those four clauses cover roughly 90% of the "I have no idea why I was banned" stories I get.

How Outlier actually detects shared accounts

The TOS would be unenforceable if Outlier couldn't see what you're doing. They can.

Here is what I can tell you they are doing, based on years of contractor experience, conversations with people who've worked on the platform side of similar operations, and the patterns I see in account-removal reports:

Device fingerprinting. Every device that logs into your account leaves a profile — operating system, browser, fonts installed, screen resolution, GPU signature, timezone, language preference. The first three or four devices that log in over your lifetime are normal. The fifteenth one, especially if it's geographically inconsistent with the first fourteen, is a flag.

IP clustering. Outlier doesn't care that you logged in from a coffee shop once. They care when two different accounts share a residential IP, log in within minutes of each other, and submit work on the same task types. That pattern shows up clearly in the backend.

Behavioral biometrics. This is the one people underestimate. Typing cadence, mouse movement patterns, task completion rhythm — these are personal enough to be effectively a signature. When your account suddenly starts behaving like a different person mid-month, the model notices. It doesn't ban you automatically, but it tags the account for review.

Quality-signal drift. If your written work was at a consistent reading level and voice for six months and then abruptly shifted, that's a flag. Not because writing styles can't evolve — they can — but because the kind of shift that happens when a different human takes over an account is statistically distinct from the kind of shift that happens when one human gets better over time.

Payment and tax cross-references. This one catches the boyfriend-girlfriend setups specifically. If the tax form on the account says one name and the Outlier internal review notices that the person describing themselves on the calibration tasks doesn't match — for instance, a chemistry PhD on file but the writing samples reflect a literature background — the discrepancy gets flagged.

I cannot tell you which one of those signals catches you. I can tell you that all of them are running, all the time, and that they don't need to be 100% accurate. They just need to be confident enough that a human reviewer agrees the account looks shared. After that, the removal is one click.

The specific behaviors that get people permabanned

Here is the list, ranked by how often I see them come up in reader emails. This is the section I'd print out and tape to your monitor if you're currently doing any of it.

1. Letting your partner or spouse log in "just to finish a task"

This is the most common cause of permabans I personally encounter. It almost always starts as a one-time thing. He's tired, she's better at writing, the task is due in an hour. She finishes it. Nothing happens. The next week, she logs in again to "see what tasks are available." The week after that, she's doing all the work and he's the name on the account.

Four to six months in, the account closes. They don't get told why. They wouldn't believe it if they were told why. But the model saw two different humans operating the account, the writing style shifted, the typing cadence shifted, the active hours shifted, and the cumulative score crossed the threshold.

I am not making a moral judgment here. I'm telling you what the system sees.

2. Working from someone else's laptop or home network

This one is subtle and unfair. You go visit your parents for a week. You bring your work laptop, but the home WiFi is spotty, so you borrow your dad's desktop for a session. You log into your Outlier account, knock out three tasks, log out.

Most of the time, nothing happens. But if your dad's IP has ever been associated with another Outlier account — say, your sibling who also tried the platform a year ago and lasted two weeks — the system now sees overlap between two accounts on the same residential IP. That's a flag.

You can do everything correctly and still get caught in the net because someone in your household made a different decision two years ago.

3. Running a second account "for your spouse"

If you and your partner both want to work on Outlier from the same household, you have to do it in a way that is genuinely two separate operations: different devices, different IPs (the system isn't that strict about residential overlap if everything else looks clean), different payment accounts, different tax IDs, and crucially, different work product that reflects each person's actual background.

Most couples who try this fail at step five. They share research, edit each other's work, copy formatting templates back and forth. The model picks it up. Both accounts close at once.

4. Using a VPN to apply from a non-supported country

Outlier supports specific countries, and the eligibility window is narrower than people realize. If you applied from a country that wasn't on the list and used a VPN to look like you were in a country that was on the list, your account is on borrowed time. The detection isn't immediate — sometimes it takes months — but tax season tends to surface it, because the paperwork doesn't match.

If this is you, the account will close. You cannot reapply legitimately from the supported country later, because the original account is permanently flagged to your payment infrastructure and tax ID.

5. AI-assist beyond what Outlier explicitly allows

The platform has its own internal tooling that may use AI. What you cannot do is run your responses through an external LLM, paste the output, and submit it. This used to be hard to detect. It is no longer hard to detect.

There are people who think the model only catches obvious GPT-style outputs. The model catches subtle patterns — sentence rhythm distributions, vocabulary frequency, hedging-language ratios. If your last three months of submissions look statistically like LLM output and your first three months didn't, that's a separate kind of flag, but it ends in the same place.

6. Creating a duplicate account after being removed

This is the cardinal sin and it doubles down on the original removal. If your first account closed for any reason — quality, TOS, inactivity, mystery — and you create a second one under a different name or email, the device fingerprint, IP, and payment infrastructure overlap will catch you within weeks. The second account closes, and now you have two ban records under your underlying identity.

I have a whole separate piece on why the reset path doesn't work the way people hope. The short version is: there is no clean reset.

What to do if you've been doing any of this

Stop. Today. Not on Monday. Not at the end of the project. Today.

If you and your partner are sharing an account, the one whose name is on it should be the only one logging in from now on. The other person needs to either get their own account from a different device and IP context, or accept that they cannot work on Outlier under the current household configuration. There isn't a third option that doesn't end in a removed account.

If you've been using a friend's laptop or a borrowed network, that needs to stop too. Going forward, your Outlier login should only happen on your own primary device, on a network you own or pay for, in the country your tax form reflects.

If you've been pasting LLM output, you already know what to do. The platform's internal tooling is the only AI you can use, and even that is bounded by the task-specific guidance.

What to do if your account has already been removed for this

You will not get it back. I want to be honest about that up front, because the appeal path for TOS-based removals is functionally closed. Outlier's support response template is the same one they use for quality removals: a polite paragraph that does not engage with the specifics, no escalation path, no human reviewer you can talk to.

What you can do is move quickly to the other platforms. Outlier doesn't share contractor records with DataAnnotation, Alignerr, Appen, Mercor, or Telus AI. Apply to all of them today, from a clean device and network if possible. Use your real name, your real tax ID, your real background. Don't try to recreate whatever workaround led to the removal at Outlier — it will catch up to you on those platforms too, and the AI gig contractor pool is small enough that getting blacklisted at three platforms is more isolating than people realize.

I wrote a more detailed runbook for the immediate aftermath of an Outlier removal — the steps to take in the first 48 hours, what to do about payments still pending, and which platforms have the shortest path back to active work. If you're newly removed, that's the next thing to read.

The honest reframe

I want to close with something that is going to feel cynical, but it's the actual truth.

Outlier is not your employer. Outlier is a platform that pays you per task, contains exactly zero of the protections of employment, and operates a detection system that is far more sophisticated than the application process suggests. The same thing is true of DataAnnotation and Alignerr and Mercor — every platform in this space is built on the assumption that some percentage of contractors will try to defraud the system, and the cost of that fraud is borne entirely by the contractors who get caught, including the ones who only nominally meant to.

The "we share an account, what's the big deal" framing is one I sympathize with as a human being and have no sympathy for as a person who watches the data. The big deal is that the platform sees it, the platform models it, and the platform removes accounts for it without a phone call. There is no version of the gig economy where this changes. There is only the version where you understand the rules well enough to stay inside them, and the version where you find out the hard way.

If you're tempted to share, don't. If you've been sharing, stop. If you've been removed, the next account in this space starts the same way every account starts: under your own name, on your own device, with your own work. That's the only setup the model isn't actively trying to catch.

I'm going to keep writing about what these platforms actually do versus what they say they do, because the gap between the two is where most of the painful surprises live. If the platform is in a strange mood this week — and Outlier just moved back into Warning status as of this morning — that gap is wider than usual, and worth paying attention to.