The law is about to arrive

This page is research, not legal advice. Every date and threshold below should be re-verified with counsel before it goes into a contract or a board paper.

With that said: the hardest date on this sector's calendar is 2 December 2026, about three months away.

#The Directive is not a proposal, it is a deadline

Directive (EU) 2024/2831 was adopted 23 October 2024 and entered into force 1 December 2024. Article 29(1) requires Member States to bring the implementing law into force by 2 December 2026 (EUR-Lex CELEX:32024L2831). Four articles do the work, and each one closes a specific escape route that operators in this atlas currently rely on.

Art. 2(1)(a) — the scope trigger is your algorithm. A "digital labour platform" is a service provided at least partly at a distance by electronic means, at a recipient's request, involving as a necessary and essential component the organisation of work performed by individuals for payment, and involving the use of automated monitoring or automated decision-making systems. A crowd platform that routes tasks by score, ranks contributors, or deactivates on a quality metric meets all four limbs. Only the fourth has any wiggle room — a purely manual ops model is arguably out of scope — and that is an untested design strategy, not a plan.

Art. 2(1)(b) — the hook is where the worker sits. "Platform work" is work "performed in the Union." Incorporation is irrelevant. A Berlin GmbH with a wholly non-EU crowd is largely outside it; a Delaware Inc. with Polish annotators is inside it.

Art. 5 — the burden of proof flips. Employment is "legally presumed… where facts indicating direction and control… are found," and "it shall be for the digital labour platform to prove that the contractual relationship… is not an employment relationship." Art. 5(6) applies the presumption prospectively from 2 December 2026 for contracts already running — so existing engagements are not retroactively converted, but they are converted going forward.

Art. 3 — the BPO workaround closes. Member States must ensure workers contracted through an intermediary get the same protection, "including, where appropriate, joint and several liability systems." Subcontracting the crowd to a local vendor stops being a firewall and becomes a shared defendant.

Alongside those: Art. 4(2) makes status turn on actual performance "irrespective of how the relationship is designated"; Art. 9 and Art. 11 require written reasons for every account restriction, suspension, termination and refusal to pay; Art. 20 requires private worker-to-worker channels the platform must not monitor.

Transposition is uneven and late. Spain (2021) and Belgium (2023) already had presumptions; several Member States had not published draft bills as of August 2026 (Littler, 20 August 2026) [WEAK — law-firm client alert]. Uneven transposition is not relief. It means twenty-seven different presumptions rather than one, and it makes where you recruit a first-order commercial decision.

#The US picture is the opposite shape: quiet federally, loud in the courts

The federal layer has three levels pointing in different directions.

The 2024 DOL rule is still on the books — a six-factor economic-reality test with no factor pre-weighted (DOL). Field Assistance Bulletin 2025-1 (1 May 2025) told investigators WHD "will no longer apply the 2024 Rule's analysis" in FLSA investigations, reverting to Fact Sheet #13 (2008) and Opinion Letter FLSA2019-6. But the FAB expressly preserves the rule: "the 2024 Rule remains in effect for purposes of private litigation" (FAB 2025-1). Enforcement stopped. The cause of action did not.

A replacement was proposed on 27 February 2026 — 91 FR 9932, comments closed 28 April 2026 — restoring two weighted core factors, control and opportunity for profit or loss (Federal Register). It has not been finalised as of late August 2026.

The joint-employer NPRM matters more commercially than the IC rule. 91 FR 21878 (23 April 2026, comments closed 22 June 2026) proposes a control-based four-factor vertical joint-employment test: who hires and fires, who supervises schedule and conditions "to a substantial degree," who sets rate and method of payment, who keeps employment records (Federal Register). That rule decides whether your lab customer gets named as a co-defendant in your crowd's wage suit — which is precisely why your customer's counsel will demand an uncapped indemnity for worker-classification claims. The rule is not an abstraction. It is a term in your MSA.

#The Kenya lesson: the intermediary shield does not hold

The Kenyan Court of Appeal ruled in September 2024 that Meta can be sued in Kenyan courts over the layoffs of Sama-employed content moderators [WEAK — Business & Human Rights Resource Centre, 23 September 2024]. Two moderator cases were still awaiting rulings in February 2026, and a judge summoned Meta executives in April 2026 [WEAK — BHRRC; Capital FM Africa]. Parallel litigation opened in Ghana in April 2025 [WEAK — The Guardian].

The structural point is not about Kenya. A court accepted jurisdiction over the foreign principal despite an in-country BPO sitting between them — the same move Art. 3 codifies. Two independent legal systems have now reached the same answer about subcontracted crowds: the layers do not insulate.

#What is actually being litigated is not what you would guess

Docket-verified, from CourtListener's RECAP search:

[UNVERIFIED — the complaints themselves could not be read; nature-of-suit codes are the only evidence of what these allege.]

The modal claim against an AI-data platform in 2026 is not wage-and-hour. It is four theories: misclassification, psychological injury from harmful content (the Kenyan moderator theory transplanted to California), fraud and non-payment over deactivations and rejected work, and trade-secret fights between platforms over poached staff and poached crowd. A wage-and-hour compliance programme addresses one of the four.

Secondary reporting adds that Scale "agreed to settle multiple lawsuits from its California contractors" [WEAK — Business Insider, 17 October 2025], and that a $12.5M wage-and-hour class settlement was being publicised in July 2026 [WEAK — settlement-administrator source].

#The ASC 606 trap: the same facts, two conclusions

Here is the part that gets missed, and it is the reason this page sits in the atlas rather than in a compliance binder.

The principal-versus-agent test under ASC 606 is control: do you control the service before it transfers to the customer? Indicators are primary responsibility for fulfilment, inventory risk, and discretion in establishing price [UNVERIFIED — codification text not retrieved]. Upwork runs both models inside one company: Marketplace is net, as agent, because it "does not control talent services" and "talent and clients negotiate and agree upon the scope and the price… directly with each other"; Enterprise Managed Services is gross, as principal, because "Lifted directly invoices the client and assumes responsibility for the work performed" (Upwork FY2025 10-K).

An operator that writes the spec, directs the work, QCs it and bears the rework risk is a principal. Revenue is gross, crowd payments are COGS, headline revenue looks four to six times larger — which is why gross versus net is the atlas's founding correction.

The consequence is a margin question, not a footnote. Contrary judges that reclassification would make Mercor's ~35% take "untenable" once benefits, overtime and compliance load are added (Contrary) — against a leaked gross margin of 27% in 2025, 33% in Q2 2026 (The Information, via BigGo). There is not much room in that number for an employment cost base.

#Where the record is thin