Understanding AI Refusal Scripts in Serbia’s Regulatory Gray Zones

Why AI Refusal Scripts Matter for Serbian Businesses

Businesses in Serbia deploying AI chatbots and voice agents navigate a complex regulatory environment. Serbia currently lacks a dedicated AI law, but existing regulations like the 2019 Personal Data Protection Act (PDPA), 2023 Ethical Guidelines for Trustworthy AI, the Consumer Protection Law, and the Advertising Act impose important limits. AI refusal scripts—predefined responses where AI declines to provide certain information or make decisions—are essential for compliance and legal risk mitigation.

Refusal scripts guide businesses through "gray zones" where AI outputs could have legal or contractual impacts, especially in fields like consumer law, medical advertising, and construction warranties. These scripts act as temporary safeguards until Serbia’s AI law is introduced by 2027, aligning with ethical standards and EU rules.

Regulatory Framework for AI in Serbia

Serbia’s AI ecosystem is indirectly regulated by several laws:

• The PDPA limits automated decisions with legal or significant effects unless explicit consent or legal authorization is present, requiring human oversight.
• The Ethical Guidelines (2023) highlight four core principles: Explainability/Verifiability, Human Dignity, Non-Maleficence, and Fairness.
• The Consumer Protection Law and Advertising Act govern claims and advertising practices, including mandatory prior consent for direct marketing.
• The EU AI Act has extraterritorial effect, impacting Serbian companies serving EU residents.

Until Serbia’s AI Law is enacted in 2027, refusal scripts provide a practical compliance tool for AI systems.

Key Areas Requiring AI Refusal to Provide Certain Information

1. Consumer Law

AI chatbots must not issue binding legal interpretations or approvals that constitute automated decisions with legal effects. For instance, a chatbot should avoid claims like, "Under Serbian consumer law you are entitled to a full refund," as this may represent a legally binding decision without human intervention.

Example refusal: "I’m unable to provide legal advice on your claim. Please contact our customer service representative for assistance."

2. Medical Advertising

AI voice agents and chatbots must not provide diagnostic, therapeutic, or drug efficacy claims. The Non-Maleficence principle from the Ethical Guidelines requires avoiding any information that might be interpreted as medical advice.

Example refusal: "I cannot provide medical diagnoses or treatment recommendations. Please consult a licensed healthcare professional."

3. Construction Warranties

AI systems must not interpret warranty coverages, liability periods, or defect scopes. Confirmations such as "Your 10-year guarantee covers this leak" risk creating automated decisions with legal effects, which is prohibited.

Example refusal: "I’m not authorized to interpret construction warranties. Please speak with our warranty department for detailed information."

Compliance Challenges for Serbian SMEs

Even businesses operating solely in Serbia must consider the EU AI Act’s extraterritorial reach, especially if they serve EU tourists, cross-border e-commerce buyers, or remote EU clients. Dual-track refusal scripts addressing both Serbian and EU regulations help manage risks related to high-risk AI outputs.

Bilingual refusal scripts must carefully align terminology in Serbian and English to avoid inconsistencies that could breach consent and profiling rules under the Advertising Act and PDPA.

Designing Effective Refusal Scripts

The Ethical Guidelines emphasize explainability, meaning refusal scripts should not merely say "I cannot answer that." Instead, they should transparently explain why AI refuses to respond, respecting human dignity and building trust.

Example of an explainable refusal script: "I am not authorized to interpret warranty terms under Serbian consumer law; I will connect you with our legal officer who can assist you further."

This approach meets human oversight requirements and prevents automated decisions with legal effects.

Common Misconceptions About AI Refusal Scripts in Serbia

• "No AI law means refusal scripts are optional." Reality: Existing laws and ethical guidelines impose enforceable limits requiring refusal scripts to avoid unlawful automated decisions.
• "Simply stating ‘I am an AI’ fulfills transparency." Reality: Transparency requires verifiability and human intervention, not just disclaimers.
• "Refusing customers harms competitiveness." Reality: Proper refusal scripts build trust and prevent liability from harmful AI outputs.
• "EU AI Act doesn’t apply if we don’t target the EU." Reality: The Act applies to any AI affecting EU residents, including Serbian companies.
• "General educational answers are always safe." Reality: Even general information can be interpreted as professional advice or legal decisions in regulated sectors.

Conclusion

AI refusal scripts in Serbia are crucial interim tools enabling businesses to comply with current data protection, consumer, and advertising laws while respecting ethical AI principles. They prevent AI systems from making unauthorized automated decisions, especially in sensitive areas like consumer claims, medical advertising, and construction warranties.

With Serbia’s AI Law expected by 2027 and ongoing EU AI Act obligations, well-crafted refusal scripts ensure businesses remain compliant and trustworthy today. Clear, transparent refusals explaining AI limitations and referring users to human experts protect both consumers and companies operating in Serbia’s evolving AI regulatory landscape.

---

Sources:

• Chambers Practice Guides on Serbia Data Protection & AI
• Zunic Law on Serbian AI Law
• Pricapartners on AI Regulation in Serbia