Pseudonymize your documents, and include the values in your AI deliverablesPseudonymize your documents consistently, and restore (de-pseudonymize) the original values in your AI assistants' outputs
Pseudonymize your documents, and include the values in your AI deliverablesPseudonymize your documents consistently, and restore (de-pseudonymize) the original values in your AI assistants' outputs
Marvin Systems CEO
Since the launch of ChatGPT in late 2022 and the subsequent widespread adoption of generative AI, two camps have been locked in a public debate.
One predicts massive job losses, the collapse of the labor market, and an unprecedented social crisis. The other heralds spectacular productivity gains, new professions, and unprecedented growth.
Neither is entirely right, or rather, both are asking the wrong question. It is undeniable that AI will transform labor markets. But the question everyone should be trying to answer is whether our democracies have the institutional capacity to manage a transition that is already underway, without eroding the social contract that underpins democratic consent.
In other words: are our political systems capable of managing these major societal changes without citizens losing confidence in the system?
This is precisely the question raised by the Carnegie Endowment for International Peace in one of its recent articles. Drawing on findings from the German Institute for Employment Research, it reveals that AI-driven disruption will likely not take the form of a sudden, mass layoff.
Instead, it will take the form of a gradual substitution of tasks, a silent automation that will first strip jobs of their substance before eliminating them. Positions will not disappear overnight. Their scope will shrink, and responsibilities will fade away insidiously. And a sense of insecurity will set in, prolonged and pervasive.
In 2025, the German Institute for Employment Research estimated that 1.6 million jobs could be transformed or eliminated by AI in Germany alone over the next fifteen years. And it is precisely the insidious nature of this disruption that makes the policy response so difficult: there is no clear moment of crisis, no deadline, no severance package to negotiate.
The finding by the German Institute for Employment Research comes as no surprise. What is more surprising is who this silent disruption will primarily affect.
Women are nearly twice as likely as men to hold jobs highly exposed to generative AI. This is confirmed by the International Labor Organization report published in March 2026.
29% of female-dominated occupations are at risk from generative AI, compared to 16% of male-dominated occupations. And when it comes to the risk of automation, the gap becomes staggering: 16% of female-dominated occupations fall into the most at-risk categories, compared to just 3% for male-dominated occupations.
The answer is more structural than cyclical.
Women have historically been concentrated in administrative, secretarial, and support roles: secretaries, receptionists, payroll clerks, and accounting assistants. These are jobs where tasks are routine, codifiable, and therefore easily replaceable by generative AI.
Men, on the other hand, are more heavily represented in construction, manufacturing, and manual trades. These are sectors where physical tasks remain, for now, more complex to automate.
It is not AI that creates this inequality. It acts more as a catalyst, revealing it and amplifying it.
Generative AI is not entering a neutral labor market. It is entering a market already shaped by decades of wage, occupational, and decision-making inequalities.
And when women are absent from the development of models and decision-making processes, the technology reproduces, and amplifies, existing biases. AI systems have, in fact, already demonstrated a tendency to disadvantage women in recruitment, compensation, and creditworthiness assessments.
This last point deserves closer attention. Credit scoring, that is, the score assigned by an algorithm to assess a person’s ability to repay a loan, is trained on historical data. Data that reflects a past in which women earned less, worked part-time more often, and had careers interrupted by maternity leave.
The algorithm does not discriminate intentionally. It reproduces patterns from the past with the appearance of mathematical objectivity. This makes discrimination even harder to challenge and therefore more dangerous.
In light of this assessment, the Carnegie Endowment article calls for a dedicated European framework for the transition of work, ideally incorporated into the EU’s 2028–2034 budget, based on three key pillars.
Redesigning social protections. Current systems were designed for specific, visible crises and are not prepared to handle the slow, insidious disruption that lies ahead. Europe needs portable social rights, applicable in all Member States, with rapid access to reskilling programs.
Establishing a continuing education infrastructure. Productivity gains linked to AI do not materialize automatically; they depend on investments in human capital. Professional expertise will remain valuable, but increasingly in combination with AI skills. Professionals must be trained on a massive scale, urgently. There is no time to lose.
Restore institutional trust. This transition is taking place against a backdrop of growing polarization and distrust of institutions. If citizens perceive that the benefits go to the few while the losses fall on the many, democratic legitimacy will erode.
The three proposed areas of focus are necessary but not sufficient.
Reforming social protections is a first step. But if no one ensures that the tools themselves are designed without reproducing the inequalities of the past, we will be building better protections for a system that still discriminates.
The impact of AI on women’s jobs is not predetermined. The International Labor Organization has a clear stance on the matter: with the right policies, serious social dialogue, and a design that respects gender equality, it is possible to prevent technology from reinforcing existing discrimination.
But this requires addressing the problem at its root: in model design, in training data, in decision-making processes, and not just in safety nets deployed after the fact.
Technology rooted in the past, deployed without safeguards, in an already unequal market.
The dice seem to be loaded before the game has even begun. And as things stand, there’s no indication that this will correct itself.
For further information : https://carnegieendowment.org/europe/strategic-europe/2026/02/how-europe-can-survive-the-ai-labor-transition