The context
AI has become a defining feature of the global economy and of daily life alike. Frontier AI models are a palpable reality, delivering sophisticated performance across highly complex tasks and industries.
This transformation is proceeding at remarkable speed but also to the detriment of values that are fundamental in any rule-of-law society. Governments worldwide are therefore attempting to reconcile the imperative of AI-driven economic growth with the demands of social safeguarding and ethical governance.
Recent projections estimate that by 2030, approximately 70% of companies will have adopted at least one AI technology, generating an additional 1.2% of annual global GDP growth. In macroeconomic terms, this is a seismic shift. Economies not structurally positioned to capture a share of that growth are, in effect, bound to fall behind, as capital will flow naturally to infrastructures capable of supporting deep technological development.
The legal and policy state of play
At the EU level, the comprehensive Artificial Intelligence for Europe strategy (COM(2018) 237) – developed through multiple coordinated frameworks since 2018 – seeks to position Europe as a global leader by advancing technological and industrial capacity, preparing society for AI-driven transformation, and establishing a robust ethical and legal governance structure consonant with EU values. This strategy is reinforced by an array of structural pillars directed at strengthening the EU’s technological capacity, ensuring an adequate ethical and legal framework, and pursuing a unified European approach.
Skills development has a key role in this process. It encompasses training opportunities, national strategies, vocational adaptation, closer co-operation between higher education and research institutions, growth in STEM graduates, and improved conditions for investment in human capital and AI literacy across member states (see the European Skills Agenda).
At the regulatory level, the AI Act (Regulation (EU) 2024/1689), in force since August 1 2024, constitutes the world’s first comprehensive legal framework for AI, employing a risk-based approach with phased implementation. It is complemented by the Digital Services Act (Regulation (EU) 2022/2065), the Data Act (Regulation (EU) 2023/2854), and the NIS2 Directive (Directive (EU) 2022/2555). Practical compliance frameworks are additionally provided by the AI Office, the AI Act Service Desk, the General-Purpose AI Code of Practice, and multiple European Commission guidelines.
The recent Digital Simplification Package (politically agreed on May 7 2026) proposes targeted amendments, including the prohibition of non-consensual intimate content generation, the reinforcement of AI Office oversight, and extended support for SMEs and regulatory sandboxes.
Member states have progressively intensified their own AI policy efforts, evolving from initial strategic frameworks towards comprehensive regulatory approaches that balance innovation incentives with trust-building, ethical rigour, and meaningful stakeholder engagement.
Romania, just like other EU member states, is developing its own regulatory and standardisation frameworks, technology adoption pathways suited to national conditions, and collaborative structures engaging academic institutions, private enterprise, and public administration.
In 2024, Romania adopted its National Strategy for Artificial Intelligence (SN-IA) 2024–2027, aiming to integrate AI technologies into the economy and society whilst upholding human rights, promoting excellence, and fostering public trust.
The strategy aligns with European and international frameworks and seeks to establish Romania as a regional hub for AI expertise and innovation. It nonetheless frankly acknowledges several significant challenges:
Insufficient qualified human resources in AI;
Limited access to adequate hardware infrastructure;
The absence of a comprehensive legal and ethical AI framework; and
Weak mechanisms for transferring research outcomes into commercial application.
Challenges and risks
AI deployment in a legal office is not primarily a technology project. It is a governance, risk, and professional responsibility project that happens to involve technology. The most resilient frameworks treat AI adoption through the simultaneous lens of quality, legal ethics, and client fiduciary duty.
Given their strategic role within the judiciary, law firms face several additional constraints. AI systems deployed in such organisations must satisfy demanding functional requirements:
Data rights (including suitable liability management in the case of AI-generated advice);
Data protection and security (meeting both GDPR obligations and legal professional privilege, including client confidentiality and professional secrecy);
Automated conflict-of-interest management;
KYC compliance;
AML compliance; or
Rigorous archiving standards.
Technological risks are equally high: hallucination and factual inaccuracy, training data bias, outdated knowledge bases, vendor lock-in, integration failures, cybersecurity vulnerabilities, or opacity in algorithmic reasoning. These are compounded by human and organisational risks – automation bias, deskilling, accountability diffusion, resistance to adoption – and by ethical and reputational concerns, including client trust and fairness, or the risk of unauthorised practice.
Governance failures represent systemic risk that individual tool performance cannot remedy. Law firms must therefore pursue ISO/IEC 42001:2023 certification and address AI risk at the level of the management system as a whole, by, for example:
Defining AI policy and objectives;
Keeping an inventory of deployed systems;
Conducting impact assessments and vendor due diligence;
Establishing human oversight mechanisms and incident response procedures;
Managing the AI system life cycle; and
Maintaining audit trails.
Unfortunately, the regulatory landscape – including AI Act implementing acts, general-purpose AI (GPAI) codes of practice, and national bar guidance – is still being crystallised, generating compliance uncertainty.
Despite these challenges, AI is already being deployed in local law firms across a wide spectrum of legal functions:
Document and knowledge management;
Litigation support;
Client-facing services (intake, triage, status portals, and AI-powered legal platforms);
Regulatory and compliance advisory work; and
Practice management (billing, resource allocation, workflow automation).
Active intervention from professional bodies
In line with the European Commission’s Communication “Ensuring Justice in the EU – a European Judicial Training Strategy for 2021–2024”, the Council of Bars and Law Societies of Europe and the European Lawyers Foundation, with the support of national bars, launched the TRADICIL project (Training of Lawyers on Digitalisation of Justice and EU Civil Law). Its objectives include training at least 500 lawyers from 26 EU member states in the digitalisation of justice.
At national level, the response of the National Union of Romanian Bar Associations (UNBR) has been notably engaged. On May 5 2026, under the auspices of the presidential administration, representatives of Romania’s major judicial institutions – including UNBR – signed an inter-institutional memorandum on the use of AI in the justice system. The memorandum confirms that AI will serve an essentially auxiliary role, with decision-making authority remaining exclusively with human professionals, and establishes a concrete governance framework comprising permanent inter-institutional working groups, pilot projects, periodic evaluations, and phased implementation.
The UNBR president identified the central challenge not as whether to use AI – already an inevitability – but whether legal professionals can maintain genuine intellectual control over it, which essentially is a matter of professional education and responsibility.
Also, ahead of the 2026 Bar Congress, UNBR published responses to the questions submitted by lawyers nationwide on the theme of reserved activities in the era of professional insecurity. Over 20 substantive responses have been offered to date, addressing, inter alia, the following:
Defining legal advice versus legal information – legal advice arises when a legal norm is applied to a specific factual situation and an individualised solution is formulated, regardless of whether the answer comes from a lawyer or an AI system. UNBR considers the creation of an explicit statutory definition of ‘legal advice’ a legislative priority, essential not to protect the profession but to activate liability, confidentiality, and independence guarantees.
Liability for AI-generated harm – for lawyers, responsibility is clear (professional liability, mandatory insurance, disciplinary mechanisms). For unauthorised platforms, liability is diffuse. Directive (EU) 2024/2853 on liability of defective products classifies AI software as a ‘product’ and places liability on the operator who integrates and deploys it commercially.
Mandatory professional standards for AI use – UNBR is developing standards on five non-delegable principles:
Legal decisions cannot be delegated to algorithms;
Human oversight must be substantive rather than formal;
The lawyer retains full liability;
Confidentiality cannot be compromised; and
AI use must be transparent where it influences the service.
Classification of AI-generated responses – UNBR has identified four platform typologies, ranging from legally compliant AI tools for legal professionals (Harvey AI, Luminance) to potentially illegal GPAI tools offering individualised legal answers to the public (ChatGPT, Gemini, etc.).
Enforcement against unauthorised practice – UNBR’s GL5 Working Group has so far identified 319 entities and 31 organisations suspected of unauthorised legal practice, obtained several criminal convictions, and is pursuing recognition as a relevant authority under the Digital Services Act to request rapid takedown of illegal online content.
Commercial branding and professional advertising – the use of terms such as ‘barou’, ‘avocatAI’, or ‘consultanță juridică’ by non-lawyer entities is treated as misleading. And on May 5 2026, the World Intellectual Property Organization ruled in UNBR’s favour, ordering the discarding of the domain unbr.eu.
Transparency obligations for AI platforms – UNBR supports minimum mandatory disclosures for AI platforms offering legal-appearing content, and confirms that it can regulate the conditions under which lawyers participate in such platforms. According to UNBR, digital transition must not be treated as a threat to lawyers but as a supportive strategy designed to eliminate bureaucratic tasks, optimise administrative efficiency, and foster internal professional dialogue.
Alternative models and ‘lawyer-bots’ – digital platforms and AI tools are legitimate forms of modernisation but cannot substitute the lawyer in providing personalised legal advice. The essential distinction is between technology that assists the lawyer and technology that purports to replace the lawyer. Mandatory AI literacy training is considered inevitable. The National Institute for the Training of Lawyers already offers an optional course on AI ethics in legal practice.
Consolidation of lawyers’ monopoly in the face of digital platforms – modern attorneys must be defined not only by their ability to litigate but increasingly by their capacity to practise preventive law and mitigate legal conflicts before they reach court.
UNBR’s overarching position is that technology cannot be prohibited but must be integrated within a regulated framework. The profession’s objective is not the defence of a professional monopoly but the preservation of the guarantees – competence, independence, confidentiality, loyalty, and accountability – that only a lawyer can provide and that no algorithm can replicate.
Quo vadis?
Pursuant to Article 113 of the AI Act, August 2 2026 triggers the application of most of the provisions not yet in force, including comprehensive requirements for high-risk AI systems listed in Annex III, spanning risk management, data governance, technical documentation, record-keeping, transparency, human oversight, accuracy, robustness, and cybersecurity, together with deployer obligations, conformity assessment procedures, and post-market monitoring. Transparency obligations under Article 50 – requiring the disclosure of AI interactions, labelling of synthetic content, and deepfake identification – also become enforceable at that date.
The Digital Omnibus will, apparently, afford businesses considerably more time to adapt in respect of certain categories: 16 months for high-risk AI in sensitive areas (Annex III) and 12 months for high-risk AI systems embedded in products covered by EU sectoral safety legislation (Annex I). Formal adoption by the European Parliament and Council is expected by July 2026; until that point, the August 2026 deadline remains technically operative.
The critical question remains: are law firm AI systems ‘high risk’? AI systems used to assist in judicial or quasi-judicial decision-making, litigation risk assessment, or case outcome prediction are the clearest candidates for Annex III classification. Strong candidates for high-risk designation further include AI tools used in proceedings before courts or administrative tribunals, AI-assisted due diligence tools determining access to legal aid, and tools used to score counterparty risk in contract negotiations. Internal document drafting assistants, research tools, billing automation, and client communication tools that do not determine outcomes affecting individuals’ rights would generally fall outside Annex III, triggering at most specific transparency obligations as per Article 50.
In practice, many law firms will operate a mixed portfolio: certain tools genuinely qualifying as high risk, and many more triggering only transparency duties.
In this complex legislative environment, it is to be expected that a significant number of firms will choose to postpone, or even decline, the deployment of AI tools – which may sit uneasily with the ambitious targets of both EU and Romanian national strategies.
