Scale AI's $12.5 million settlement prices what "Orwellian" annotation monitoring costs in California
A California court has given preliminary approval to a $12.5 million settlement resolving claims that Scale AI misclassified its Contributors, the gig workers who annotate images, write example chatbot responses and rate model outputs, as independent contractors instead of employees. The complaint, filed in San Francisco Superior Court in December 2024, described Scale's monitoring of workers as "Orwellian": screenshot-capturing software, pay docked when a task ran longer than Scale's own time estimate, and silent reassignment to lower-paying projects. The final approval hearing is set for 30 October 2026.
What changed, and when
McKinney, et al. v. Scale AI, Inc., et al., Case No. CGC-24-620481, Superior Court of California, County of San Francisco, was filed in December 2024 by lead plaintiff Steve McKinney, a former Contributor. Defendants include Scale AI, Smart Ecosystem, and staffing firm HireArt. Contributors, per the settlement notice, "annotate images, write example responses, and rate chatbot outputs" through platforms including Outlier and Remotasks.
The settlement class covers anyone who worked as a Contributor for Scale, Smart Ecosystem, or through HireArt while residing in California from 10 December 2020 through 28 February 2026. The deadline to opt out, object, or challenge a workweek count was 3 September 2026. The final approval hearing is scheduled for 30 October 2026, 9:00 AM, Department 304, San Francisco Superior Court.
The evidence table
| Item | Amount / date | Source |
|---|---|---|
| Gross settlement amount | $12,500,000 | settlement notice |
| Attorneys' fees (cap, one third) | $4,166,666.67 | settlement notice |
| Litigation costs (cap) | $100,000 | settlement notice |
| Settlement administrator (cap) | $100,000 | settlement notice |
| PAGA penalties (cap) | $500,000 | settlement notice |
| (of which to California LWDA, 65%) | $325,000 | settlement notice |
| (of which to PAGA workers directly, 35%) | $175,000 | settlement notice |
| Estimated net to class | $7,543,333.33 | settlement notice (estimate) |
| Number of workers in the class | not reported | not disclosed in settlement notice |
| Case filed | December 2024 | settlement notice, case docket |
| Class period | 10 Dec 2020 to 28 Feb 2026 | settlement notice |
| Opt-out / objection deadline | 3 September 2026 | settlement notice |
| Final approval hearing | 30 October 2026 | settlement notice |
Source: openclassactions.com settlement notice for McKinney v. Scale AI, Inc., CGC-24-620481, San Francisco Superior Court, corroborated on case background by kingsiegel.com. Neither source is the court docket itself; the settlement amount, cap figures and the two 2026 dates are consistent across both.
How piece-rate annotation work becomes a misclassification claim
The legal mechanism is California's ABC test for independent-contractor status (codified by AB 5 and tested in Dynamex), under which a worker is presumed an employee unless the hiring company proves the work is outside its usual course of business, is performed without the company's direction and control, and is customarily done as an independently established trade. Platform-mediated annotation work, where the platform assigns tasks, sets pay rates and time estimates, and monitors output, sits awkwardly against the "without direction and control" prong, which is the opening the complaint used.
The specific allegations map onto that prong directly. Screenshot-monitoring software is direction and control over how work is performed, not just what is delivered. Docking pay when a task exceeds Scale's own time estimate ties pay to the company's performance standard rather than a rate the worker negotiated. Reassignment to lower-paying projects "without notice" is the company unilaterally controlling the work itself, not responding to a contractor's own availability or choice. None of these three allegations individually proves misclassification, but together they describe exactly the kind of operational control the ABC test asks a company to disprove.
What it means for anyone running or buying contractor-based annotation labor
A concrete settlement number now exists for benchmarking exposure. $12.5 million, over a class period spanning roughly five years and two months, resolving claims against one of the largest annotation and RLHF data vendors, is a real reference point for legal reserve planning at any company structuring similar work through 1099 contractors in California, not just Scale AI.
The alleged practices are a checklist of what increases exposure, not a list unique to one company. Output-monitoring software, pay rates tied unilaterally to the platform's own time estimates, and task reassignment without the worker's input are common features of piece-rate annotation platforms generally. This settlement does not establish that any of them is independently illegal, it resolves specific claims without an adjudicated finding of liability, but it shows what a plaintiff's complaint built around them can be worth.
California's ABC test applies regardless of the work's subject matter. Nothing about the misclassification claim turns on the work being AI training data specifically; the same legal theory applies to any gig-platform labor in California. What makes this case notable for the annotation industry is simply that annotation and RLHF labeling is now one of the largest sources of platform-mediated gig work the test is being applied to.
Our judgement: this is a settlement, not a verdict, reached without any court ruling on whether the underlying allegations are true. Both of our sources are secondary (a settlement-tracking site and a plaintiff-side law firm blog), not the court docket itself, so the exact preliminary-approval date is unconfirmed, though the settlement amount and the two 2026 dates are consistent across both.
Check it yourself
The case is a matter of public record in San Francisco Superior Court.
open https://openclassactions.com/settlements/employment-law/scale-ai-contributor-misclassification-class-action-settlement.php
open https://www.kingsiegel.com/blog/scale-ai-misclassification-settlement-california-class-action/
# case lookup (San Francisco Superior Court public case access)
# Case Number: CGC-24-620481
# Case Name: McKinney, et al. v. Scale AI, Inc., et al.
# Court: Superior Court of California, County of San Francisco
# Department: 304 (final approval hearing)
python3 - <<'PY'
gross = 12_500_000
fees = 4_166_666.67
costs = 100_000
admin = 100_000
paga = 500_000
net_est = 7_543_333.33
accounted = fees + costs + admin + paga + net_est
print("sum of disclosed line items: $%.2f" % accounted)
print("gross minus sum of line items: $%.2f (rounding in the notice's own estimate)" % (gross - accounted))
PY
# sum of disclosed line items: $12410000.00
# gross minus sum of line items: $90000.00 (rounding in the notice's own estimate)The roughly $90,000 gap between the gross amount and the sum of disclosed line items is in the settlement notice's own published estimate, not a calculation error on our part; we report it rather than silently reconciling it, since the notice does not explain the gap.
What would prove this wrong
A settlement is a negotiated resolution, not a judicial finding, so there is no "finding" here to falsify in the way a scientific claim could be. What can be checked is whether this is actually resolved as described.
A dated prediction: by 30 November 2026, one month after the scheduled final approval hearing, the court record for CGC-24-620481 will show either final approval of the settlement substantially as described here, or a continuance, objection, or amendment. If the final hearing is postponed past that window or the terms change materially, the specific dates and amounts in this piece should be treated as superseded, not as the final outcome.
Sources
- Open Class Actions. Scale AI Contributor Misclassification Class Action Settlement, McKinney, et al. v. Scale AI, Inc., et al., Case No. CGC-24-620481, San Francisco Superior Court.
- King Siegel LLP. Scale AI Misclassification Settlement: California Class Action, case background and allegations.
- BLOMEGA. Unpaid annotation tasks in ACL papers, 2018 to 2025.
- BLOMEGA. Consented AI training data providers.
- BLOMEGA. Data annotation research: the latest.