Meta's Automation Push Triggers Strike at Dublin Firm Covalen
720 Covalen workers face redundancy as Meta automates Irish annotation. CWU voted for strikes over terms leaving 400 workers with nothing.
DUBLIN β Workers at Covalen, the Irish outsourcing firm that provides AI annotation and content moderation to Meta, voted on May 7 in favor of strike action, with walkouts scheduled for May 15 and 22. The dispute follows Covalen's announcement that 720 jobs on Meta-related projects are at risk β more than 500 of them in AI annotation β and centers on a redundancy package the Communications Workers' Union (CWU) calls well below the Irish tech sector norm.
Workers also staged a protest outside the DΓ‘il to demand government intervention, marking one of the most visible organized labor actions the AI training industry has yet seen.
The Redundancy Dispute
Covalen is offering only the Irish statutory minimum: two weeks' pay per year of service, capped at β¬600 per week, and only for workers who have completed two full years of employment. The CWU argues this falls well short of what other major outsourcing firms operating in Ireland β including Accenture and Genpact β have provided in comparable situations.
CWU organiser John Bohan noted that AI annotation workers at Covalen earn average salaries of around β¬32,000 and that statutory-only redundancy leaves many, particularly migrant workers, in an extremely precarious position given Ireland's cost of living and thin social security net. "Statutory redundancy only is absolutely an abnormal and anti-worker move in an Irish context, especially so in large, profitable businesses with major clients," Bohan said.
CPL Resources, Covalen's parent company, recorded a profit of β¬26.1 million in 2024, up from β¬23.6 million the year before. Union members have demanded to know what compensation Covalen receives from Meta when contracts are reduced, arguing that workers have the right to understand what resources exist to mitigate job losses.
The Two-Year Cliff
The eligibility threshold is where the dispute becomes most acute. Under Irish law, statutory redundancy requires two complete years of service. According to an open letter from a 20-member committee elected by CWU members, roughly 400 of the 720 affected workers will receive no payment at all. Among them: 58 workers who will miss the two-year qualifying threshold by only a few weeks.
Workers are also contesting the timeline of the statutory consultation period. Covalen has argued the 30-day consultation window began on April 27, when staff were informed of layoffs via a video call. The union disputes this, noting that employee representatives did not meet with management until a week after that call β meaning the legal clock, in their view, has not yet properly started. If the matter is not resolved, the CWU has indicated it will escalate to the Workplace Relations Commission and, if necessary, the Labour Court.
Workers have also flagged a strict six-month "cooling-off" clause that prevents laid-off Covalen staff from taking roles with other Meta-contracted firms β effectively locking them out of their most likely re-employment market for half a year.
Why Meta Is Cutting
Meta confirmed in March that it is "deploying more advanced AI systems" to handle content enforcement internally, which is directly reducing its reliance on third-party vendors like Covalen. The company has announced plans to cut roughly 8,000 jobs globally β 10% of its workforce β as it redirects tens of billions of dollars toward AI infrastructure in 2026.
This is not isolated to Covalen. Meta began this vendor reduction wave in January 2026, when roughly 300 Covalen workers were laid off in an earlier round. Workers at the time also took strike action. The May announcement of 720 further redundancies is the second major wave in under five months.
The concern raised by workers and safety researchers is that AI is not yet reliable enough to replace human judgment in content moderation. Those posting illegal content β including child sexual abuse material β are often adept at disguising its format to evade algorithmic detection. The CWU and UNI Global Union have both warned that discarding experienced human moderators in favor of automation carries genuine safety risks that Meta has not publicly addressed.
The Intermediary Shield
A March 2026 report by SOMO (Centre for Research on Multinational Corporations) mapped the structural arrangement behind this kind of layoff. The report identified what it called an "intermediary shield" β a layered network of at least 30 intermediary companies that tech giants like Meta use to scale AI training and moderation operations while insulating themselves from direct labor liability when priorities shift.
Under this model, the platform company sets the business direction (including the decision to automate), while contractual risk and workforce costs sit with vendors like Covalen. When Meta moves work in-house, Covalen carries the redundancy burden β not Meta. UNI Global Union General Secretary Christy Hoffman called this out directly: "Tech companies are treating the workers whose labour and data helped build AI as disposable."
Ian McArdle, Deputy General Secretary of the CWU, framed the structural dynamic plainly: "Meta is shedding thousands of jobs worldwide, cutting staff and tearing up vendor contracts simply to pay for its massive new Artificial Intelligence bills. Once again, tech workers are losing their livelihoods to fund the shifting priorities of tech giants."
What This Means for AI Trainers
The Covalen situation is a high-signal indicator of where the broader AI training labor market is heading. Several trends are converging:
Generalist annotation is being automated or insourced. Basic content moderation and simple labeling β the work that defines most vendor-side annotation roles β is increasingly handled by internal AI systems or synthetic data pipelines. Vendor headcount in this tier will continue to fall as model capabilities improve.
Specialist roles remain in demand. Demand for high-expertise RLHF work β legal, medical, coding, and adversarial red-teaming β remains strong and is harder to automate or move in-house. The divergence between these two tiers is accelerating.
Labor is organizing. The Covalen strike is among the first significant collective actions in the AI training sector. Coming alongside the April letter from 40 US labor groups to Congress and UNI Global Union's public statement, it signals a shift: this workforce is no longer invisible. Workers are building institutional capacity to push back.
Forward Outlook
Covalen has said it remains committed to "direct and meaningful dialogue" but had not officially received notice of industrial action as of May 7. Strikes are scheduled for May 15 and 22 unless negotiations improve.
Three things are worth watching. First, whether the Irish government responds to the DΓ‘il protest with concrete legislative proposals β the CWU is demanding regulatory protections for tech-sector workers facing AI-driven layoffs, not just assessments. Second, whether Meta faces reputational or regulatory pressure in Europe over its handling of vendor workforces under the EU AI Act's supply-chain provisions. Third, whether the Covalen dispute serves as a template for similar actions at other Meta vendors across the UK, continental Europe, and the Philippines, where comparable annotation workforces exist.
For workers in the AI training industry more broadly, the message from Dublin is unambiguous: the transition from human-powered annotation to internal automation is happening faster than the labor protections designed to manage that transition. Organizing β formally and early β is the only countermeasure that has shown any leverage so far.
Sources
- The Journal β "Covalen workers vote to strike after 720 workers on Meta projects told jobs at risk" (May 7, 2026)
- UNI Global Union β "UNI calls on Meta to take responsibility for workers in its supply chain" (April 29, 2026)
- SOMO β "Intermediary Shield" report on Big Tech vendor labor structures (March 2026)
Related reading
Is AI training legit? β how to identify platforms with transparent labor practices before you apply.
Running multiple platforms at once β how to reduce dependence on any single vendor or client.
Best AI training platforms compared β ranked alternatives to single-vendor annotation work.
What is data labeling? β the category of work at the center of the Covalen dispute.

Pietro R.
MSc Human-Computer Interaction | Founder & Product Owner
Pietro is the founder and technical lead of aitrainer.work. He builds and maintains the platform's data pipeline, certification infrastructure, and editorial standards.