AI at Work: the enforcement tide reshaping employers’ choices
Regulators, courts and workers are forcing employers and vendors to answer for opaque algorithms, surveillance and disclosure — and HR must pick sides.
This week the simplest way to read the pile of AI stories is as one signal: enforcement and pushback are no longer theoretical. Regulators in Europe are already using new powers; US state and federal agencies are rewriting disclosure and enforcement playbooks; unions and workers are mobilising; and vendors are racing to add controls, watermarks and audit trails. That combination is changing not only what HR buys, but what it must do when questions land on the desk.
Enforcement moves from rulemaking to action
The European Union has shifted from policy drafting to enforcement. The European AI Office issued its first formal information requests to model providers this week, an unmistakable step from talk to scrutiny EU AI Office info requests. At the same time the Commission has started enforcing Article 50 transparency duties, and major providers have responded — Anthropic says Claude models now carry invisible watermarks and C2PA provenance to meet those duties Anthropic embeds invisible watermark. Brussels also designated services such as ChatGPT under the DSA regime, bringing a different compliance clock for platform oversight EU designates ChatGPT, Reddit and Roblox. These moves matter for HR teams that consume third‑party models or deploy screening and agentic tools: transparency, provenance and vendor cooperation are becoming procurement line items overnight.
Complementing the EU push, national instruments and regulators are already landing enforcement-sized blows. The Dutch data protection authority fined Uber nearly €825m for fully automated driver deactivations and a separate Amsterdam claim alleges opaque algorithms set pay in breach of data‑protection rules Dutch regulator fines Uber Uber faces Europe-wide driver class action. In the UK, the ICO is finalising binding guidance on automated decision‑making while the Department for Business and Trade consults on rules for workplace monitoring ICO confirms timetable Make Work Pay consultation. The regulatory spotlight has moved from possibility to precedent, and HR contracts, DPIAs and vendor checklists must reflect that.
Surveillance, workforce trust and industrial pressure
Legal risk is meeting worker resistance. Nurses have mobilised against hospital contracts with Palantir-style scheduling systems, while unions from SIPTU to the TUC are demanding statutory consultation, impact assessments and retraining funds when employers introduce AI Nurses protest Palantir ties TUC urges statutory union say SIPTU ballots public-service action. Surveys show the ground is shaky: almost half of workers say they had no voice in rollouts and a large majority fear over‑dependency on AI tools Remesh: 43% ACC: 74% of US workers worry.
At the coalface, employers are already collating detailed dossiers from location traces, message scraping and camera feeds; those practices are sparking community and legal backlash — from counties cancelling license‑plate camera contracts in the US to venues worrying about covert recording with smartglasses in the UK Employers building employee dossiers Flock Safety faces Texas backlash Ray‑Ban glasses covert recording. For HR, the lesson is blunt: surveillance choices are talent choices. Poorly scoped monitoring erodes trust and invites union and regulatory responses that are already costly to manage.
Litigation is sharpening the rules employers thought they owned
Courts and class actions are converting theoretical harms into real discovery and costs. A federal judge in California has allowed disparate‑impact and age claims against Workday to proceed, signaling that vendor‑powered hiring systems will be litigated aggressively Judge allows claims against Workday. Parallel class suits target other hiring vendors and employers for opaque candidate scoring that can replicate bias Lawsuits target Eightfold AI. Employment tribunals are also wrestling with redundancy decisions tied to AI projects, as a UK tribunal found a developer unfairly dismissed in a pre‑ordained redundancy linked to an AI initiative Tribunal rules agency unfairly dismissed developer.
These cases come against a backdrop of regulatory prioritisation: the EEOC’s new strategic plan singles out algorithmic hiring and performance tools for enforcement, and NYC audits have criticised weak follow‑through on local bias‑audit obligations EEOC strategic plan NYC Local Law 144 audit. HR teams should expect more discovery requests, audit demands and an expanding checklist of documentation — from model provenance to candidate notices — when disputes arrive.
Vendors, product shifts and operational strain
Vendors are responding unevenly. Major HR platforms are accelerating AI features while attempting to cabin risk: Workday has pushed Agent Passport and deeper AWS integration even as it adds hiring triggers and ranking templates for recruiters Workday unveils Agent Passport Workday adds HiredScore triggers. Fireflies launched voice agents for screening calls, immediately raising recruitment disclosure and data‑handling questions under the EU AI Act and GDPR Fireflies launches Voice Agents. SAP added pay‑transparency analytics to help employers meet new EU reporting obligations, a sign that compliance functionality is becoming a selling point SAP adds EU Pay Transparency Insights.
At the same time, product and platform quirks are complicating technical controls: researchers found that common Windows apps embed server GUIDs into generated images, a detail that could intersect awkwardly with EU provenance and data‑protection rules Researcher finds Paint and Photos embed GUIDs. Meanwhile public bodies are considering their vendor mix: the Commission is weighing moving away from US cloud providers and particular models in its job‑matching app on sovereignty grounds EU executive weighs ditching US cloud and Anthropic. HR procurement must now ask for watermarking, C2PA manifests and stronger contractual commitments on model behaviour.
Hiring, inequality and the new skills calculus
On the demand side, AI is reshaping job markets and raising competence gaps. LinkedIn data shows AI‑labelled roles pay well, but women remain underrepresented among hires, a stark signal for inclusion strategies LinkedIn: AI job listings pay ~$177K. Stanford research warns of an "algorithmic monoculture" in hiring where common models can create position‑level racial gaps, reinforcing the need for audit and mitigation Stanford study finds job-level racial gaps. Employers are already reacting by shifting hiring mixes and seeking AI‑literate security and ops staff as agentic tools change threat profiles and operational tasks AI agents reshape IT security hiring Goldman Sachs: AI curbs entry-level hiring.
The upshot for HR is practical: talent frameworks must absorb technology risk (who oversees agents, who signs off DPIAs) and equity risk (how to detect and correct disparate outcomes) while retraining budgets and consultation processes become bargaining levers in workplaces where unions and regulators are active.
What to watch
Next week watch the European AI Office and national data‑protection authorities for follow‑through on the formal information requests and Article 50 enforcement: those actions will shape whether provenance and watermarking become standard contractual terms in HR tech procurement. The ICO’s timetable for automated‑decision guidance and the UK government’s Make Work Pay consultation will be the UK test for how far statutory controls over workplace monitoring and biometric surveillance will go.
In the US, state law shifts and litigation are the live signals. Colorado’s amendments to its AI law and California’s SB 951 — adding AI disclosures to WARN notices — show states are moving toward disclosure‑first regimes; employers planning restructures should prepare to list affected job classifications and vendor details. Finally, keep an eye on discovery in the big vendor cases and on vendor road maps: product additions that promise agentic HR automation will also carry new provenance requirements and likely new audit obligations that HR teams must operationalise immediately.