Under the General Data Protection Regulation (GDPR), consent is only one of the legal bases that can be used to process personal data, including when using Large Language Models (LLMs) to boost the efficiency of processing personal information. While obtaining consent has its advantages, for example the possibility of a clear audit trail if consent is properly managed, it is not without its own pitfalls. Valid consent must be informed, freely given, specific, and unambiguous. This is by no means a low hurdle, as we can see from hefty fines being imposed for not meeting these requirements.
When making the decision on which legal basis to rely upon when processing personal data via LLMs, it will be helpful to know in which instances consent may be the preferred or even required choice, and that’s what we explore in this article.
Examples Where Consent is the Safer Choice or Required:
Personalized Marketing Using LLMs: If a company wants to use LLMs to analyze customer behavior and create personalized marketing campaigns, explicit consent would typically be required when the individual who is marketed to is a child. Recital 38 provides that children merit special protection with regard to their personal data because they may be less aware of the risks, consequences, and safeguards as well as their rights. Furthermore, the recital explicitly brings the example of marketing and creating personality or user profiles as an instance where special protection is particularly crucial. The recital does, however, not specifically require that this special protection must be provided in the form of a strict consent requirement.
There might be other ways to grant special protection, unless of course a general consent requirement already exists for personalized marketing. That is not the case, though. Recital 47 says that “The processing of personal data for direct marketing purposes may be regarded as carried out for a legitimate interest.” Legitimate interest is an alternative legal basis which organizations can rely on for the processing of personal data. There are, however, strict requirements that must be fulfilled, first and foremost a demonstration that the business’s interests are not overridden by the interests or fundamental rights and freedoms of the data subject. Recital 47 does not negate this requirement; plausibly, it merely ensures that direct marketing is not per se excluded from being considered as a legitimate interest. Thus, it is possible for the legitimate interest basis to suffice, even when the data of children is concerned, but in order for ‘special protection’ to be granted, at the very least, the interests, fundamental rights and freedoms of the child will have to be weighted more heavily against the business interests.
A final consideration must be given to Recital 50 which addresses the processing of personal data for purposes other than those for which the data was collected. Generally, it is not required to establish an independent legal basis for such secondary purposes, provided the secondary processing purpose is “compatible” with the original purpose, the determination of which warrants its own post. Highlighting only one element that factors into the compatibility assessment, the reasonable expectations of the individual must be considered. Presumably, there will be many instances in which the data that an organization wishes to run through an LLM model to expedite personalized marketing campaigns was originally collected for a different purpose, perhaps on the legitimate interest basis. Here, Recital 50 says “Where the data subject has given consent […], the controller should be allowed to further process the personal data irrespective of the compatibility of the purposes.” In other words, if the legitimate interest exception was relied upon for the original, primary purpose collection of the data, a determination must be made whether the personalized marketing purpose is ‘compatible’ with the primary purpose. If, on the other hand, consent is obtained for personalized marketing, the onerous compatibility determination is not necessary.
Thus, for both adults’ and children’s data processing for personalized marketing purposes it is a good idea to obtain express consent. While relying on the legitimate interest exception is possible, at least in the case of adults’ data, the balancing of interests may result in the finding that the interests of individuals may override the marketing interests of organizations.
When it comes to email marketing, whether supported by LLMs or not, Article 13(1) of Directive 2002/58/EC, aka the ePrivacy Directive, prohibits direct marketing of non-customers without consent. This may change again, though, if the ePrivacy Regulation ever arises from the current stalemate between EU institutions it has been stuck in for almost 2 years.
Sentiment Analysis on Social Media Comments: Sentiment analysis, often referred to as opinion mining, is a powerful tool that allows businesses to gauge public opinion by analyzing the emotional tone behind social media posts, comments, and reviews. By leveraging Large Language Models (LLMs) and other advanced AI techniques, companies can gain insights into customer preferences, emerging trends, and potential areas of concern.
Social media content, even when publicly available, can be considered personal data if it can be used to identify an individual either directly or in conjunction with other data. Consequently, the GDPR applies, and we have to ask what requirements must be met to allow such processing of this data.
As discussed above, it is possible that the collection of opinions is the primary purpose and that a legitimate interest can form a valid basis for this collection. But if the analysis is used for purposes beyond what the user might reasonably expect, such as targeted advertising, a compatibility analysis is required, or obtaining consent for this secondary purpose.
Obtaining consent may be the better option if the opinions collected contain “special categories” of data, which include details about an individual’s race, religion, political opinions, health, and more. For instance, analyzing social media content for sentiments regarding a political event might capture individuals’ political affiliations, in which case the data cannot be processed based on legitimate interest.
Article 9 of the GDPR sets out 10 permissible reasons for processing special categories of data, consent being the one that will most plausibly apply in the case of sentiment analysis.
Examples Where Consent May Not Be Required:
Processing for Contractual Obligations: If an organization uses LLMs to process personal data for the purpose of fulfilling a contract with the individual, such as processing an online order, consent may not be required. In this case, the legal basis for processing could be the necessity to fulfill a contract. Imagine an online travel booking platform that offers flight, hotel, and car rental reservations. Customers can browse options, select their preferences, and complete the booking process through the platform’s website or mobile app. Now the platform decides to implement an LLM-powered chatbot to assist customers during the booking process. The chatbot can answer questions, provide recommendations, and even facilitate the booking process by collecting necessary information such as travel dates, destinations, passenger details, and payment information. Note that payment information is not included in the definition of special categories of personal data under Article 9, hence explicit consent for this kind of processing does not seem to be required.
Compliance with Legal Obligations: If the processing of personal data using LLMs is necessary for compliance with a legal obligation to which the organization is subject, consent may not be required. For example, if a financial institution uses LLMs to detect fraudulent activities, this could be considered a legal obligation because the bank is subject to anti-money laundering and anti-terrorist financing obligations.
Conclusion
The requirement for consent under the GDPR when using LLMs to process personal information is nuanced and depends on the specific context, purpose, and type of processed information. Organizations must carefully assess the nature of the data and the expectations of the individuals to determine whether consent is required. It is advisable to consult with legal experts or data protection authorities to ensure compliance with the GDPR, as failure to obtain proper consent when required can lead to significant fines and penalties.
