Accessibility and data security
The responsible use of AI requires the examination of critical issues concerning user protection. This section focuses on the safety of vulnerable groups, such as minors and persons with disabilities, as well as the fundamental need for the protection of personal data and confidentiality.
Protection of children and minors
The interaction of minors with LLMs entails significant risks. Uncontrolled access can expose minors to inappropriate content, misinformation, or manipulation, as these systems are not designed to recognize the vulnerability of children. The position of the Greek State, aligned with European Law, is that minors constitute a vulnerable group and their protection is of the highest priority.
Age Limits
Under 13 years old: Most public LLM platforms explicitly prohibit use by individuals under 13 years of age.
13-18 years old: Verifiable parental consent is required.
It should be noted that even special “student editions” (13-18 years old) with enhanced safety filters do not eliminate the risks and do not replace critical parental involvement.
Responsibility of Parents/Guardians
- Parental supervision is imperative, and use by children under 13 years old must be prevented.
- For teenagers (13-18 years old), use must occur following active discussion,
- the utilization of available parental control tools, and clear guidance.
- It is important for parents to educate minors about specific risks, such as “hallucinations”, bias, the sharing of personal data, and the risk of exposure to inappropriate interactions.
Accessibility and equal use for persons with disabilities
AI systems possess significant potential to enhance digital accessibility and independent living for persons with disabilities. However, the technology is still in a developmental stage regarding the assurance of universal and equal use. Many of the current LLM interfaces are not yet optimized for assistive technologies or may generate responses in formats that are not universally accessible (e.g., responses in table format).
Positive Developments
Despite the challenges, examples where AI functions as a digital assistant are already being observed. Modern mobile devices integrate AI functionalities enabling the use of LLMs by persons with visual impairments and supporting alternative forms of communication.
Data security and confidentiality
The use of publicly available LLMs raises significant data protection and confidentiality issues. Information inputted into an LLM may become accessible to third parties, and entering sensitive or personal information into public LLM platforms entails significant risks:
Risks
Data leakage: Information inputted into an LLM may become accessible to third parties through technical security flaws or by being incorporated into the LLM’s training data. Furthermore, conversations may be stored and used to improve or further train the LLMs, according to the respective provider’s terms of use.
Non-compliance with regulations: Prompts and data inputted into commercial LLMs are typically processed on third-party provider infrastructures, often outside the European Union.
Loss of control: Once data is transferred to a third party’s computing infrastructure, control over it is significantly limited.
Among others, adhering to the following protection rules is recommended:
Protection Rules
- Avoid entering sensitive personal data, such as Tax Identification Numbers (TINs), ID numbers, home addresses, phone numbers, medical data, financial information, or any other information that could identify or endanger an individual.
- Carefully read the terms of use and privacy policy of the respective LLM. Some platforms offer settings that allow restricting the storage or use of data for training purposes.
- Use the “temporary chat” feature, where available. In this mode, offered by many LLMs, conversations are not stored in the history and are not used to train the LLM.
- Data anonymization whenever feasible. Replace real names, locations, or other identifiers with generic or fictitious references.
- Select approved tools and systems, especially within the context of professional or public use. Organizations should carefully evaluate the compliance of LLMs with data protection regulations prior to their adoption.
Awareness of these limitations and implementation of the recommended practices ensure that the use of LLMs remains safe, responsible, and in accordance with the principles of ethical AI use that run through this guide.