Bulgaria - AI in Education Strategy

Overview

The Draft Strategy for the Development and Integration of Artificial Intelligence in Bulgarian Education (proposed May 2025) is a national policy instrument intended to provide a structured, rights-respecting pathway for the deployment of AI across pre‑school, school, vocational and higher education. It proposes an outcomes-focused approach that balances opportunities (personalized learning, administrative efficiency, new pedagogies) with systemic risks (privacy and child protection, bias in automated assessment, cybersecurity). The draft explicitly references and seeks operational alignment with Bulgaria’s national AI Concept and recent government-level action on AI. Official foundational documents and partnerships highlighted during the strategy’s formulation include the government’s national AI concept (Government Concept for AI to 2030) and nationally developed language-model and AI research capacity (e.g., INSAIT’s BgGPT project) that are referenced as strategic assets for education-focused deployments (INSAIT: BgGPT announcement). The overview foregrounds the draft’s dual goals: to (1) create a safe, transparent regulatory and operational context for AI in education; and (2) build capacity and infrastructure so that Bulgarian schools and higher education institutions can benefit from domestically aligned AI solutions.

Definitions

The draft proposes clear, education-specific definitions to avoid ambiguity: "Artificial Intelligence (AI) in education" covers algorithmic systems and tools that perform tasks normally requiring human cognition (e.g., language generation, predictive analytics, classification) when used in pedagogical, administrative or evaluative contexts. "High-stakes educational decision" is defined as any automated decision that materially affects student progression, certification, access to programmes, or disciplinary outcomes. "Provider" means any commercial or non-commercial entity supplying AI systems, components or services to educational institutions. "Personal data of minors" follows the GDPR definition but is treated with enhanced safeguards in the educational context. "Human-in-the-loop" indicates a design requirement where a qualified educator retains meaningful intervention and override capability for system outputs that affect learning or assessment.

Governance and Institutional Framework

The proposed governance model establishes an inter-ministerial Steering Board chaired by the Ministry of Education and Science and including representatives from the Commission for Personal Data Protection (CPDP), Ministry of Transport, Information Technology and Communications (where relevant for infrastructure), the Executive Agency for Education Programme management, and independent academic and civil-society experts. A specialized Secretariat hosted by the Ministry would coordinate implementation, maintain an approved-systems register and manage procurement guidelines. The draft calls for explicit Memoranda of Understanding with national research institutes (e.g., the Institute for Computer Science, Artificial Intelligence and Technology/INSAIT) to support safe model adaptation and auditing. It also proposes regional coordination points (via Regional Education Inspectorates) to support pilot rollouts and collect compliance data. Where commercial vendors are involved, the governance framework mandates contractual clauses covering data processing terms, audit rights, security obligations and liability allocations. To ensure transparency and stakeholder participation, the draft prescribes public consultation phases for major procurements and for updates to the national AI-in-education guidance. For international and EU policy coherence, the Secretariat is tasked with maintaining a regulatory watch function to track the EU AI Act and other binding obligations and to prepare conformity and registration pathways for identified high-risk systems (Ministry of Education and Science (sitemap and AI page)).

Key Focus Areas

The draft organizes policy and operational activity into six interdependent pillars: (1) Curriculum and competencies — integrating AI literacy (both "learning about AI" and "learning with AI") across subjects and age groups, and embedding critical thinking and digital citizenship; (2) Teacher and leader capacity — national CPD frameworks, accredited courses, and an approved-provider registry to scale high-quality training; (3) Technical infrastructure and procurement — secure connectivity, standardized APIs, local or contractual data residency and procurement templates favoring explainability and auditable systems; (4) Safe pedagogical deployments — protocols for pilot design, human oversight, student consent/notification, and limitations on algorithmic profiling for high‑stakes decisions; (5) Data protection and child safety — mandatory Data Protection Impact Assessments (DPIAs) for systems processing pupil data, pseudonymization where possible, strict retention limits and parental information processes; (6) Research, evaluation and innovation — targeted funding for research partnerships (e.g., with INSAIT), open-model initiatives, and evidence-generation on learning outcomes and equity. Each pillar has associated deliverables: updated curriculum frameworks, an accredited teacher-training catalogue, technical procurement templates and minimum-security baselines, DPIA templates, pilot evaluation protocols and an independent monitoring dashboard. The strategy also emphasizes inclusion: targeted support for rural schools, assistive AI for learners with disabilities, and language adaptation for Bulgarian and minority languages.

Implementation Framework

The draft proposes a phased implementation: Phase 0 (preparation, months 0–6) establishes governance, publishes mandatory guidance (DPIA template, procurement templates, minimum-security checklist) and opens calls for pilot partners; Phase 1 (pilots, months 6–24) funds regionally representative pilots across different school types (urban/rural, general/professional, special educational needs) with independent evaluation; Phase 2 (scale-up, months 24–60) implements validated systems, supports procurement and teacher CPD at scale, and integrates AI literacy into national curricula; Phase 3 (consolidation, months 60+) focuses on continuous improvement, long-term funding and full alignment with EU conformity regimes. The draft requires providers of medium- and high-risk systems to submit technical documentation, third-party evaluation or conformity assessment (as the EU AI Act prescribes) and to agree to ongoing post-market monitoring obligations. It also proposes a national sandbox for research and testing under tightly controlled data governance, enabling safe model fine-tuning with anonymized educational datasets and partnerships with research institutions such as INSAIT. Funding mechanisms are a mix of national budget lines and EU co-financing (Education Programme 2021–2027). Procurement templates incorporate ethics-by-design and audit clauses, and require demonstration of bias‑mitigation testing and robustness checks prior to deployment.

Monitoring and Evaluation

Monitoring is two-fold: compliance monitoring (regulatory and contractual) and impact evaluation (learning outcomes, equity and safety metrics). The Secretariat will publish an annual Compliance Report summarizing DPIAs, audits, incidents, enforcement actions and the approved-systems register. Impact evaluation will be undertaken by independent academic teams, using standardized outcome and equity indicators and pre-/post-implementation study designs. The strategy envisages a national dashboard presenting anonymized learning analytics, distributional effects by socio-economic status, incidents of harm (e.g., inappropriate content or data breaches) and system performance metrics. Post-deployment, providers must submit routine safety testing reports and update documentation when models are materially changed. The strategy calls for an early-warning incident reporting mechanism for schools to notify the Secretariat and CPDP within defined timelines should breaches or harmful outcomes occur.

Penalties, Liability, and Appeals

The draft sets out a graduated enforcement regime, combining administrative measures and contractual remedies. Proposed penalties include suspension of system use in schools, removal from the Ministry’s approved-providers list, temporary or permanent exclusion from public procurement, and financial penalties under relevant administrative law (to be coordinated with the CPDP for GDPR-related fines). For contractual breaches, remedies include funding clawbacks, termination, and claims for damages. The strategy specifies an appeals process allowing vendors and schools to request review by an independent panel (including academic experts and civil-society representatives) prior to final sanctions. It also calls for clear parental and pupil redress routes — a formal complaints process managed by the Secretariat with referral pathways to CPDP or judicial remedies where personal-data or fundamental-rights violations are alleged.

Relationship to Other Instruments

The draft situates itself within Bulgaria’s broader regulatory environment and international obligations. It explicitly references the Government Concept on AI to 2030 and aligns education-specific measures with GDPR (data protection), national child-protection laws and procurement rules. The draft anticipates the EU AI Act (conformity assessment, categorization of high-risk systems, market surveillance obligations) and proposes a national registration and market surveillance layer for educational AI systems to complement EU processes. It also references national programmes such as the Education Programme 2021–2027 (EU funding) and existing teacher CPD registers (RQT), proposing interoperability between administrative registers and the new approved-systems registry. Where sectoral rules exist (e.g., health data in school-based health services), those regimes take precedence and are integrated through inter-agency cooperation agreements.

International Alignment

The draft emphasises alignment with EU regulatory developments (notably the EU AI Act), OECD guidance on AI in education and UNESCO recommendations on AI and education. It foresees active engagement with EU digital education initiatives, peer learning through the European Education Area and technical cooperation with pan-European research infrastructures. The strategy supports open-science approaches (where feasible) and encourages exportable, ethically designed Bulgarian AI solutions, including partnerships with national research centres (INSAIT) to develop language- and curriculum‑adapted models. It also proposes mechanisms to incorporate international best practices (bias testing, child-centred design, explainability standards) into national procurement and certification templates and to contribute Bulgarian case studies to EU policy discussions.

Implementation Timeline

Sources and References