16.02.2026 | Articles, SAVALnews
AI is rapidly becoming part of our daily lives. There are more language models available than there is time to try them out, search services provide AI-generated summaries and when browsing through social media feeds, anything can pop up.
In the clearest cases, it is easy to identify and assess the reliability of AI output. But what if AI produces a convincing sounding answer to a difficult question? Or manages to draw up an agreement that seems valid?
AI text is often fluent and convincing, and the terms may be accurate or at least sound accurate. The text might even contain reliable source references. AI does not, however, understand law, at least not yet. It does not distinguish the law from provisions in a collective agreement, nor is it versed in legal praxis. Furthermore, AI does not stop to take details into account – the exceptions to the exception on which our entire legal system is built. In general, AI also tends to endeavour to please the user: by formulating a question in a different way, you might get a different answer to the same thing.
Thus, when AI is drawing up its convincing answer to a legal question, it does not pause to consider all the details of collective agreements or to look up any appropriate case law. It does not even check what the law or, in particular, the law drafting process has said about the matter. Instead, AI calculates probabilities. It simply predicts which word is most likely to follow another. If the probability calculation fails, AI will randomly select an answer, i.e. an AI hallucination. Trusting such a response can be costly for the user.
How good are the language models that are specifically trained using legal material and are intended to support the work of a lawyer? Even they don’t always give the right answers. When you point out to AI that it has made a mistake, such as an outdated law reference, it sincerely apologises and usually corrects it. That’s sweet but doesn’t inspire much trust. And, of course, that’s truly all that can be expected from AI. It is not the job of AI to take responsibility for its mistakes.
Despite the above, AI can serve as a good tool for lawyers. General language models help with translation and text parsing, while language models trained using legal material search for sources, such as relevant legal cases. Thus, it is possible to significantly increase the efficiency of the work with the help of AI. It does, however, require one to be vigilant and mindful to notice when information is out of date or completely inaccurate, and when the sources used are irrelevant.
At the moment, AI is a good tool for brainstorming, routine tasks and information retrieval. It does not, however, replace the lawyer. If a contract is signed on the basis of AI guidance, the possibility for mistakes is high. You may find yourself feeling very alone, at the latest, when a disagreement arises concerning the interpretation of an agreement drafted by artificial intelligence. Who should you turn to then?
Note: The names have been changed
Elli’s employer had proposed that instead of dismissal, her employment could be terminated through a joint agreement. In this case, Elli would receive severance pay. It was decided that a draft agreement would be drawn up using AI.
AI produced a neat and appropriate-looking document. The severance pay was also mentioned. It all seemed good at first glance. However, Elli still wanted the draft agreement to be reviewed by lawyers at ASIA and, in the process, we discovered that she had more than one hundred balance hours that remained unused. This was not mentioned in the agreement. In addition, according to the agreement, the employment relationship would terminate immediately upon signing, even if Elli’s normal period of notice would have been one month. Elli was unilaterally waiving requirements – the agreement only talked about the employee's obligation to waive requirements or claims and not a word about the employer’s corresponding obligation. When discussing the matter further with Elli, it turned out that she would also have the right to transition security allowance if dismissed.
Suddenly, the offered agreement with severance pay did not seem to be a very attractive option. By accepting the agreement, Elli would be left high and dry compared to the situation in which she was dismissed. Thus, the agreement was renegotiated on the basis of our advice and, in the end, Elli was able to sign the agreement with peace of mind.
Change negotiations had been initiated at Juha’s workplace. Juha wanted to know the minimum duration of time, by law, an employer must use to conduct such negotiations. It was a fairly straightforward question, so he decided to ask AI.
The answer was clearly laid out: the duration of such negotiations depends on the scope of the planned actions and is 14 days or six weeks. Juha discussed the matter with a shop steward at his workplace who knew that the law had changed recently and that the negotiation periods under the new law were now seven days or three weeks. Since Juha worked in the technology industry as a senior salaried employee, he fell within the scope of the sector's collective agreement. The collective agreement stated that the negotiation period for the sector is 28 days when the negotiations concern the dismissal of at least ten people.
After pointing out the error to AI, Juha also eventually got an AI response that corresponded to the latest legislation. Although Juha initially told that he was a senior salaried employee in the technology industry, the AI service knew nothing about the 28-day negotiation period stated in the collective agreement. Fortunately, the shop steward was well-versed in the facts.
Whether you are faced with a draft contract, an acute question or negotiations with your employer, it would be wise to contact the lawyers of ASIA. We know the law, collective agreements and legal praxis – and most importantly, we always consider your individual situation.
You shouldn't be surprised if our answer is exactly the opposite of the answer that AI has already given you, or if AI seems to be more certain of the issue. When it comes to law, it is typical for there to be no definitive or exhaustive answer to many issues. Even in such situations, AI seems often to be (unjustifiably) self-confident.
Text by Karen Saarinen, Labor Market Lawyer at ASIA ry. Published on ASIA member magazine 1/2026
Contact information
Become member