June 20, 2026

The week in one read: The platforms are stress-testing how much they can take from workers before workers leave — and finding out the answer is "a lot." Three platforms ran mass deactivation events this week, DataAnnotation's drought deepened into a genuine income crisis, and Handshake AI logged its worst stretch since launch. Across the board, the pattern is the same: work disappears, pay disappears, and support is either automated or absent.

DataAnnotation

The drought is real, severe, and getting worse. A 1.5-year veteran called it the worst and longest they've seen — task volume collapsed from ~$2K/month averages to $600 pace for June, with EU and non-US workers near zero. The geographic split is sharp: US workers in core project pools are doing okay; everyone else is hitting F5 for nothing. Two separate deactivation events this week, including one worker cut mid-task after 14 months and $18,565 earned. The adversarial coding tasks (7–11 hours, no guaranteed pay if you can't break the model) are still running, and many workers didn't know you could submit for time without achieving a failure — if that's you, read the docs.

"Last year I averaged roughly $2k a month. This month I'm on pace for $600."

Handshake AI

Complete breakdown. Time-tracking changed to mouse-movement-based idle detection and is silently eating hours — workers who thought they logged a full week are finding 17 hours on the dashboard. A mass Slack lockout hit this week, widely read as a covert deactivation wave timed to avoid paying recently submitted hours. Payment shortfalls are ranging from 1 hour to 61 hours per person, with multiple workers reporting rent-level sums missing. The R2I assessment deployed June 17th is failing workers who felt confident, with no score, no feedback, and no appeal — community speculation is pointing at a planned 20–35% workforce reduction using the assessment as cover. If you're on Handshake, assume nothing until it hits your account.

"I worked 77 and they paid me for 16 oh fuck no i gotta complain"

Outlier AI

The Aether project ran a mandatory screener and then mass-deactivated people who completed it — including oracle-rated workers with months of tenure, 4.2+ ratings, and no AI use. Multiple workers report same-day permanent bans, no meaningful human review, and no appeal path. A prior mass ban event two months ago was eventually reversed as a system bug; workers are watching to see if that happens again, but support is auto-closing tickets. If you're on Aether and you see a Center Circle or S2S screening prompt, understand what you may be walking into.

"8 months of work just to randomly deactivate people's accounts"

Mercor

Two things to flag. First: a coordinated scam operation is using Mercor's assessment infrastructure as a credibility prop — fake employers request assessments, collect passport and SSN during the process, then ghost. This is documented and ongoing. Don't hand identity documents to anyone outside official Mercor channels. Second: the 23K-person project is running a one-strike offboarding rule on a new 4-hour task type, and 800 workers were removed in a single day this week. The "fight-for-your-life hustle" framing from workers inside it is not hyperbole. Mid-project switches from hourly to pay-per-task are also being reported as deliberate and recurring.

"They turned a really good project into a fight-for-your-life hustle."

Alignerr

Pattern becoming clearer: contracts are being terminated after tasks pass QC but before earnings convert to payroll, leaving workers with accepted work and zero pay. Multiple independent accounts confirm this, including at least one worker who completed an unpaid assessment, had tasks approved, and then had their contract cut. The "fast-tracked $90/hr" email that workers are receiving is leading into post-eval silence, not actual onboarding. Don't invest significant unpaid time here until the pay structure is verifiable.

"I guess if they kick you off the project before your completed tasks gets converted to payroll then you don't get paid. Wild, right?"

Stellar AI

The LWI project disappeared mid-task for affected workers — forum gone, 404 errors, no communication. The A1 project completed a qualification round, briefly relaunched under a new name, and went back down within a day. Workers who passed quals have no tasks and no timeline. Stellar's pattern of qualifying large cohorts and then pausing before any real work ships is becoming a recognizable cycle.

Telus AI

Task windows are running 10–60 minutes across most locales before going NTA for hours. India workers are being pushed to overnight availability to find work at all. The tryrating platform is also crashing intermittently, compounding the lost time. The platform requires 10-hour weekly minimums to maintain standing but isn't consistently providing 10 hours of tasks.

"If we're expected to work 10 hours to keep our job when there are tasks, we should ALWAYS have a min of 10 hours or be paid 10 hours even if there aren't tasks!"

Watch next week: Whether Outlier reverses the Aether deactivation wave as a "system bug" (as it did ~2 months ago), and whether Handshake's R2I lockout translates into confirmed non-payment for the hours already submitted.