When an employment decision starts with AI-generated content

There is a name for what is happening to much of the content published on social platforms. Researchers have been calling it the falsification of specialised knowledge. People with no experience in a given subject use artificial intelligence tools to write the script and then present themselves as an authority on the topic.

The scale is already significant. A survey reported in August 2026 indicates that one in five videos shown to a new YouTube user is low-quality AI-generated content. The platform itself acknowledges the risk and lists health, finance and legal matters among the subjects where it weighs most heavily.

For anyone running a company, the problem does not stay on the screen. This content reaches the HR department, reaches the line manager and, at some point, becomes a decision about an employment contract.

Why this is a legal issue and not merely a cultural one

An employment decision is not an opinion. It takes immediate effect on the contract and, in practice, is almost always irreversible.

A dismissal for cause is not undone by an apology. A working time offset arrangement put in place without the proper instrument contaminates months of records. A hiring structure adopted outside the correct model persists for the entire duration of the relationship.

Article 2 of the Brazilian Labor Code assigns the risks of the economic activity to the employer. Whoever hires, pays and directs the personal provision of services is the party accountable for what was decided, regardless of where the information behind that decision came from. There is no defence in labour proceedings based on the source the manager consulted.

Three recurring situations

Dismissal for cause applied for conduct the law does not cover. Article 482 of the CLT sets out an exhaustive list of grounds, and each requires its own elements, evidence and, for some of them, a graduated scale of penalties. Content that reduces the provision to a list of punishable behaviours tends to omit precisely the requirements that decide the case, such as promptness of the penalty, proportionality and the prohibition on punishing the same fact twice.

Working time offset arrangements set up without the required instrument. Article 59 of the CLT provides different instruments depending on the offset period. An individual written agreement and a collective bargaining instrument are not interchangeable, and the offset period determines which is required. General material on hour banking rarely draws this distinction, and the consequence of getting it wrong is the invalidity of the arrangement, with the hours payable as overtime.

Engagement as a legal entity adopted without concrete analysis. The lawfulness of a civil engagement depends on the reality of the arrangement, not on the label of the contract. The analysis runs through the elements of article 3 of the CLT, above all subordination. Content announcing that engagement through a legal entity is now permitted treats as a general rule something that depends entirely on the specific case.

What artificial intelligence does well, and what it does not do

There is nothing wrong with using artificial intelligence in legal and HR routines. It is excellent at organising information, producing a first draft, comparing texts and freeing up qualified working time. This firm uses these tools daily.

What it does not do is take on the decision.

The tool does not know the collective bargaining agreement applicable to that category, has not read the employee's disciplinary record, does not know whether a prior warning was issued for the same fact, does not weigh the available evidence and is accountable for nothing. It produces a plausible text, and plausibility is precisely what makes the error hard to notice.

The risk grows because the format increases apparent credibility. Synthetic information delivered on video, with narration and the rhythm of a lecture, is received as though it came from someone who has mastered the subject.

The loss arrives with a characteristic delay

A flawed decision produces no visible effect on the day it is taken. It generates no immediate claim, appears in no management indicator and draws no attention.

The effect surfaces when the labour claim is filed, frequently two years later, and the company discovers it never held the document that would have supported that decision. By then, the missing evidence can no longer be produced, because it needed to be contemporaneous with the facts.

That lag is what makes the problem hard for management to see. By the time the cost appears, no one connects the liability to a decision taken years earlier.

Governance that does not add bureaucracy

The aim is not to prohibit the tool, but to place technical responsibility between the answer and the decision:

• Define which acts require prior legal review. Dismissal for cause, changes to working time, changes to the engagement model, transfers, reductions in hours and pay, and termination of employees with job security or on medical leave.
• Record the source of the guidance. When a decision is revisited two years later, knowing what the manager relied on shortens the investigation considerably.
• Maintain validated internal templates, so the team does not have to look outside for what the company should already be providing.
• Treat the collective bargaining agreement as a mandatory reference. Much of the content in circulation ignores that a category may have its own, more restrictive rule.
• Train the team on the tool's limits, focusing on which questions it answers well and which require analysis of the specific case.

The firm's practice

Reviewing HR policies and decision workflows, providing prior review of measures that alter employment contracts, drafting internal templates and training HR teams form part of the firm's corporate employment law practice. Each case, however, depends on its own circumstances, facts and documents.

Conclusion

The growth of AI-generated content is not, in itself, a legal problem. The problem begins when that content replaces technical analysis in decisions that alter employment contracts.

No tool takes on the decision. It remains the company's, and the company is the party that answers for it and bears the consequences.

A good tool speeds up someone who knows what they are doing, and speeds up with equal efficiency the mistake of someone who does not. Before applying any measure that touches an employment contract, two questions to the team resolve much of the problem. Where did this information come from, and who is accountable for it.

Reference basis for review

• CLT, article 2 (the employer bears the risks of the economic activity).
• CLT, article 3 (elements characterising an employment relationship).
• CLT, article 59 and its paragraphs (working time offset and hour banking, with different instruments depending on the offset period).
• CLT, article 482 (grounds for dismissal for cause, an exhaustive list).
• BBC News Brasil report published on 12 August 2026 on AI-generated channels, recording research by the company Kapwing according to which twenty per cent of the content shown to a new YouTube user is low-quality AI-generated video, along with the platform's position on sensitive subjects.
• Study by Shuo Niu, associate professor of Computer Science at Clark University, published in March 2026, analysing 377 videos from 47 English-language channels dedicated to teaching how to generate income with artificial intelligence, and the source of the expression falsification of specialised knowledge.
• Brazilian Bar Association Rule 205/2021, to preserve the informational character of the content.

Informational content only. This does not constitute legal advice, an offer of services or a promise of results. Any concrete assessment depends on the facts, the documents and the context of each case.