What is a large language model?
A large language model (LLM) is a neural network — today almost always a transformer — trained on massive text corpora to predict the next token. At sufficient scale, that single objective yields the ability to draft, summarise, translate, code and reason across domains. LLMs are the dominant family of general-purpose AI models and the substrate of chatbots, copilots and AI agents.
From prediction to product
How a company touches an LLM shapes its legal position: calling a model through an API, fine-tuning it on proprietary data and training one from scratch sit on very different rungs of the responsibility ladder. Most Turkish products sit on the first rung; the analysis below still applies at every rung, because input, output and model-layer questions arise at each of them.
The legal profile of an LLM
- Inputs: training corpora raise copyright questions — the TDM opt-out regime in the EU — and personal-data questions wherever the corpus contains personal data;
- Outputs: hallucination makes accuracy claims risky; AI-generated content triggers Article 50 marking obligations from August 2026; ownership of outputs runs through copyright analysis (FSEK in Türkiye);
- Model layer: providers placing LLMs on the EU market carry Article 53 GPAI duties — technical documentation, a copyright policy and a public training-content summary.
Turkish context
Türkiye has no TDM exception: FSEK (5846) contains no text-and-data-mining carve-out, so training on scraped Turkish content cannot lean on an EU-style exception. KVKK (6698) applies equally to personal data in corpora and in prompts, and no AI-specific statute is in force. Turkish startups building on foundation models mostly stay downstream of Article 53, but Article 50 marking and KVKK duties follow the product wherever it ships in the EU.
Do: treat model choice as a legal decision — record what the provider discloses about training data and limitations before you build on it. Don’t: promise factual accuracy in customer-facing copy; hallucination is a documented property of the technology, and your terms should state what the model is and is not warranted to do.
Related guides: Web Scraping for AI in Turkish Law, Training Data and KVKK.
Working on this? Vircon Legal advises on AI & Algorithm Law and AI Compliance Hub. Talk to us →
If this is on your desk
Templates and checklists are free in the Founder Academy; for a specific situation, book a 30-minute intro call.
Founder AcademyBook an intro call