1. Executive summary
The Vienna-based digital rights group noyb filed five complaints on 27 October 2026 against Scale AI in France, Italy, the Netherlands and Austria, alleging that the company's Outlier data-labelling platform tracks freelance workers' web activity without a valid legal basis and requires biometric verification as a condition of paid work. The complaints name 18 third-party trackers identified on French freelancers' sessions, including TikTok, LinkedIn, Facebook, Sprig and Datadog, and allege that Scale collects IP addresses, contact details, unique identifiers and visited-page data, in some cases recording mouse clicks and keystrokes every few seconds. Scale disputes the claims and says screen recording is opt-in and biometric checks exist to combat fraud. No CVE, exploitation activity or threat actor is associated with this item; the risk to EMEA financial services is regulatory and third-party, not technical compromise. The exposure for clients is concentrated in two places: any use of crowd-sourced or vendor-supplied labelling platforms that embed third-party trackers, and any outsourced identity-verification flow (here Persona) that processes biometric data of EU residents.
2. Regulatory framing
| Article | Trigger (the fact in this item) | Practical impact |
|---|---|---|
| DORA Art. 28 | Scale AI's Outlier platform is used by Google, Meta and OpenAI to generate model training data, and the complaints allege third-party trackers (TikTok, LinkedIn, Facebook, Sprig, Datadog) embedded in worker sessions | A firm that contracts a labelling or annotation provider inherits the provider's sub-processor and tracker chain; Art. 28 requires the firm to assess and contract for that ICT third-party risk, including the provider's own dependencies |
| DORA Art. 30 | The tracker and biometric-verification allegations concern how a contracted platform processes data and which sub-processors (Persona, Datadog, Sprig) it engages | Contractual provisions with the ICT provider must cover sub-contracting, data location and audit rights; a provider that cannot answer a data-subject access request (as alleged here) is a contractual gap the firm must close |
| NIS2 Art. 21(2)(d) | The item describes a supplier relationship in which the supplier's own supply chain (third-party trackers, an external identity verifier) is the source of the alleged exposure | Supply chain security measures must extend to the supplier's sub-processors, not just the direct contract; firms relying on such platforms need evidence of the sub-processor inventory |
No DORA or NIS2 incident-reporting obligation is engaged: the item describes alleged unlawful processing, not an ICT-related incident affecting the firm's own services.
3. Technical analysis & attack chain
This is a regulatory and privacy action, not a cyber intrusion. There is no exploitation chain, no malware and no attacker. The relevant "chain" is the data flow noyb alleges, and it is set out below as the complainants describe it. Every element is contested by Scale and none has been adjudicated.
- Scale contracts freelance workers through the Outlier platform to generate AI model training data. Scale says it has worked with more than 900,000 people; Meta owns nearly half of Scale.
- Workers are required to complete identity verification through Persona, submitting a face video and identification documents. noyb characterises this as a condition of earning, which Scale disputes, saying verification is used to combat fraud and misuse and to comply with European law.
- While workers perform tasks, noyb alleges the platform records mouse clicks and keystrokes at intervals of a few seconds, and that the collected data could be assembled into video documenting an entire workday. Scale says it records screen activity and task duration only where workers opt in.
- noyb's analysis identified 18 third-party trackers on French freelancers' sessions, including TikTok, LinkedIn, Facebook, Sprig and Datadog. A noyb spokesperson said the count varies by country and is not exhaustive because the analysis focused on social media and analytics trackers.
- The data noyb says the trackers collect includes IP addresses, contact details, unique identifiers and information about the visited page, and it says this can support targeted advertising.
- Four freelancers requested copies of all personal data held from both the tracking system and the biometric verification process for the period December 2025 to May 2026. noyb says Scale did not respond.
The technical specifics that matter to a defender are the named components, not any exploit: Persona as the identity-verification processor, and Datadog, Sprig, TikTok, LinkedIn and Facebook as the tracker vendors noyb identified. The tracker count and the characterisation of the click and keystroke recording rest on noyb's own analysis and press release; they are single-sourced and Scale disputes them. The data-subject access request non-response is also single-sourced to noyb. No independent technical review of the Outlier client, its scripts or its network calls has been published in the source material, so the tracker inventory should be treated as indicative rather than complete.
4. Mitigation & containment
P1 (within 24 hours)
- Identify every contract under which the firm sends work to a crowd-sourced or vendor-hosted labelling, annotation or data-preparation platform, and confirm whether that platform is Outlier or a reseller of it. Record the contracting entity, the data categories sent and the sub-processors named in the contract.
- For any such contract, request the provider's current sub-processor list and confirm whether Persona, Datadog, Sprig, TikTok, LinkedIn or Facebook appear. A provider that cannot produce the list is a DORA Art. 28 exposure and should be escalated to procurement and the outsourcing register.
P2 (within 72 hours)
- Where the firm's own staff or contractors use a labelling platform, review the client-side scripts and network calls the platform makes from firm-managed devices. Block tracker domains at the web proxy or DNS layer for firm devices if the platform's terms permit, and log the blocks.
- Review any biometric onboarding flow that uses Persona or a comparable verifier. Confirm the lawful basis, the retention period and the deletion-request process, and test the data-subject access request path end to end. The allegation that a provider failed to answer four access requests over a six-month window is the failure mode to test for.
P3 (within 7 days)
- Update the third-party risk assessment for any AI data-services supplier to record the tracker and identity-verification sub-processors, and add contractual language requiring notification of sub-processor changes and cooperation with data-subject requests (DORA Art. 30).
- Where the firm supplies data to a labelling provider rather than using its workers, confirm the provider does not reuse firm data for model training. Scale states it does not use contributors' personal data to train customer models; obtain that in writing from any provider the firm uses.
No patch, version pin or vendor fix applies. There is no CVE and no software remediation in this item.
5. Indicators of compromise
No indicators of compromise available in the source material. The item names commercial services (Persona, Datadog, Sprig, TikTok, LinkedIn, Facebook) as alleged data recipients, but gives no domains, URLs, hashes or network artefacts, and none should be inferred.
Behavioural indicators
| behaviour | where to observe | confidence |
|---|---|---|
| Third-party tracker scripts loading during work sessions on a labelling platform | Browser or endpoint telemetry on worker devices; web proxy logs for requests to social media and analytics domains during platform use | Low; single-sourced to noyb, count and identity disputed by Scale |
| Keystroke and mouse-click capture at multi-second intervals during task execution | Endpoint input-instrumentation telemetry, if the platform client is instrumented | Low; single-sourced to noyb, Scale says recording is opt-in |
| Biometric identity verification (face video plus ID document) required before task allocation | Identity-verification vendor logs (Persona); onboarding workflow records | Medium; both parties agree verification occurs, they dispute whether it is mandatory |
| Data-subject access requests unanswered beyond statutory deadline | Privacy operations ticketing; provider correspondence | Low; single-sourced to noyb |
6. Detection
Insufficient indicators to author detection rules.
The source contains no file artefacts, command lines, registry keys, process names or distinctive strings belonging to a threat. The named services are legitimate commercial products, and a rule that matched on their names would detect normal business traffic, not malicious activity.
7. Sources
- The Record, "European digital rights nonprofit noyb sues Scale AI for 'invasive' worker surveillance", Suzanne Smalley, 2026-10-28, https://therecord.media/noyb-sues-scale-ai-for-worker-surveillance
This is the only source in the item. All claims about noyb's allegations, the tracker count, the biometric requirement and Scale's response rest on this single report and the statements it quotes. No primary court filing, noyb press release or Scale statement was reviewed directly.
8. Adverse Trace position
Adverse Trace assesses this as a privacy and third-party risk item with no technical exploitation component and no threat actor; the CVSS, KEV and EPSS fields do not apply because no CVE is associated with it. The impact on EMEA financial services is indirect: firms that contract AI data-labelling or annotation providers inherit the provider's tracker and identity-verification sub-processor chain, and the allegations here describe exactly the failure mode DORA Art. 28 and Art. 30 are written to catch, namely a supplier that cannot account for its own sub-processors or answer a data-subject request. The tracker count, the keystroke-recording claim and the unanswered access requests are single-sourced to noyb and disputed by Scale, so they should be treated as allegations to test against a firm's own contracts rather than established fact. We will monitor the five complaints in France, Italy, the Netherlands and Austria for regulatory outcomes and any supervisory authority action, and we will update this note if a primary filing or a Scale response with sub-processor detail becomes available. Clients who use Outlier or a comparable platform should expect to be asked for their sub-processor inventory and their data-subject request evidence during the next outsourcing review.
Published via PulseTrace — Adverse Trace threat intelligence.