The First Employee Dismissed by Artificial Intelligence
On 19 November 2025, the Rome Court filed a judgment that, given the right headline, seems straight out of a 1980s science-fiction film: a worker dismissed because artificial intelligence had replaced her. Judgment No. 9135 attracted extensive media and legal attention, with coverage by Corriere, Repubblica, Sky and Il Sole 24 Ore, and lengthy discussion among employment lawyers and trade unions. For a few weeks on LinkedIn, it seemed Skynet had found its own trusted employment lawyer.
Unfortunately for the headline, a careful reading reveals something far less cinematic.
The case concerned a graphic designer employed by a cybersecurity company. The business was experiencing documented financial difficulties—corporate changes, terminated commercial relationships, even eviction for unpaid rent—and decided to reorganise around its core business: software development and cyber intelligence. Its graphic design department was abolished. Some remaining tasks were absorbed by a manager, who performed them with assistance from AI tools.
The employee challenged her dismissal. The Court rejected her claim.
This is where media accounts stop, and where the problem begins. The judge did not say that artificial intelligence had dismissed somebody. The judgment describes AI as one tool within an otherwise genuine and proven reorganisation. This is not a semantic detail. It is the entire difference between a viral headline and a properly understood judgment.
For a dismissal on justified objective grounds under Article 3 of Italian Law No. 604/1966 to be lawful, three requirements must be satisfied: the organisational decision must be genuine and not a pretext; it must cause the elimination of the position; and redeployment must be impossible. Financial crisis is not necessary: such a dismissal may also result from reorganisation aimed simply at productive efficiency—a phrase that sounds neutral in a contract but must be explained word by word in court. In the Rome case, a crisis existed and was particularly important as evidence. It is not a general legal requirement, but a particularly strong way of proving one.
In short, AI did not dismiss anyone. It simply made it more credible to the judge that the position had genuinely become redundant.
This is precisely the same legal framework applied decades ago when management software entered accounting departments and made staff who completed ledgers by hand redundant. The tool changes. The legal category that must accommodate it does not.
But here lies the more interesting issue, one the Rome judgment did not expressly address because it did not need to: what happens when AI replaces not one graphic designer but reduces staffing across an entire department?
An indirect contribution came five months later. In Order No. 9668 of 15 April 2026, the Italian Supreme Court of Cassation—dealing with an entirely different case, the restructuring of a care home—reaffirmed a principle immediately relevant to any AI-assisted reorganisation: when redundancies affect a homogeneous group of interchangeable workers, proving the need to reduce headcount is not enough. The employer must also justify why that particular employee was selected, using criteria consistent with fairness and good faith and an adequate comparison of all potential candidates at the same level.
Put plainly, if a company automates an entire administrative department or first-line customer support team, it cannot simply say “that role is no longer needed”. It must show why one employee rather than another was dismissed when duties and grade were equivalent. That evidential burden is far from trivial. The next Italian legal battle is likely to be fought here, in the next AI-assisted reduction of a homogeneous group of employees.
Looking beyond Italy helps show how open the issue remains. On 28 April 2026, the Hangzhou Intermediate People’s Court—in one of China’s AI hubs, making the case almost a domestic dress rehearsal—publicised a comparable case that reached the opposite conclusion. Zhou, a quality-control supervisor at a technology company, checked the accuracy of language-model outputs for a monthly salary of 25,000 yuan, approximately €3,200. He spent his days correcting the work of a machine that would soon be deemed good enough to dispense with him. The company proposed transferring him to an ordinary operational role and reducing his salary to 15,000 yuan, approximately €1,920. Zhou refused and was dismissed.
Only when Zhou brought the dispute to arbitration did the company reveal its real explanation: as the models improved, much of the checking could now be done by AI itself.
The dismissal was held unlawful in arbitration and at both first instance and appeal. Zhou was awarded over 260,000 yuan, approximately €33,300, under the double-compensation rule in Chinese law. The reasoning was that voluntarily introducing AI to cut costs does not automatically constitute the “substantial change in objective circumstances” required by Chinese law to justify termination—a category reserved for events such as mergers or relocation. The company was not in difficulty. It had simply chosen to save money, and discovered that saving money badly can be surprisingly expensive.
Read together, Rome and Hangzhou tell a subtler story than a simple East–West divide. Both legal systems ultimately say the same thing: AI alone is not an autonomous legal ground for dismissal. But they place the point of scrutiny differently. Chinese law in this case also examined the economic rationale underlying the business decision, distinguishing objective necessity from mere convenience. Italian law does not: the judge cannot review the economic expediency of the organisational choice, only whether it was genuinely implemented and was not a pretext. In Italy, you may choose a machine for reasons that do not concern the judge at all. You must simply be ready to prove you actually made that choice, rather than merely said you did.
An Italian judge will ultimately ask not whether the business was entitled to prefer a machine to an employee, but whether the reorganisation was genuinely implemented, whether it actually eliminated that position, whether the employee could have worked elsewhere, and, where several employees were interchangeable, why that particular person was selected.
AI-assisted reorganisations are no longer hypothetical: they already appear in this year’s business plans.
The interesting question is not whether companies can use AI to reduce their need for human labour. They can.
The question is whether, when litigation follows, they will be able to prove that the reorganisation was genuine, the position really disappeared, no redeployment was possible and—if several equivalent employees existed—why that particular person lost their job.
Because an algorithm can make an employee redundant.
It does not make evidence redundant.
Paolo Fortina · Originally published on LinkedIn on 5 September 2026. Read the original
Comments