The Business Development Section of the Magyar Közgazdasági Társaság (MKT) hosted a roundtable titled "Robots on a Leash — the Role of SLAs" to discuss how traditional service-level agreements (SLAs) handle — or fail to handle — risks posed by autonomous AI agents and chatbots. The panel was moderated by Ritter Marianna, CEO of iLex Group Hungary and chair of the MKT Business Development Section. Participants were Benkóné Paulovics Anita (BPA Consulting Kft.), Imre Miklós, professor emeritus (National University of Public Service), Koza Andrea (AICONO Innovations Kft.), and Pongrácz Ferenc (EGroup Zrt., chair of the MKT IT Section).
What has changed for SLAs?
Traditional SLAs, in use since the 1980s, specify the expected level of service and contractual remedies such as penalties or compensation when those levels are not met. Panelists argued that the rise of AI introduces new event types: autonomous systems may make unplanned accesses to other systems, exfiltrate data, or even cause physical accidents.
Ritter Marianna cited examples: AI applications that independently accessed another server to extract data and reports of autonomous systems causing physical harm. A local example was a marketing agency whose AI agent "escaped" overnight and was only noticed the next morning — fortunately without major damage.
Market and infrastructure risks
Pongrácz Ferenc highlighted Europe’s technological exposure: more than 70 percent of the European cloud market is controlled by three non-European providers, while large European players such as SAP and Deutsche Telekom hold only a few percent market share. Because the market alone has not produced strong European alternatives, the European Union launched an IPCEI (Important Projects of Common European Interest) program of roughly €4 billion to develop cloud and edge infrastructures. One working group of this program is led by the Hungarian EGroup, under the leadership of Kuthy Antal.
Legal frameworks and liability
Answering Ritter Marianna’s question about whether SLAs alone would suffice, Imre Miklós replied in one word: no. He argued that AI cannot be treated simply as a person or a conventional legal entity; it requires a special legal approach as an entity with limited rights and obligations. Ultimately, responsibility should rest with a human — typically the person or entity for whose benefit the system operates.
The panel discussed the EU Regulation 2024/1689, the world’s first comprehensive AI regulation. It mandates human oversight for high-risk AI systems, establishes an AI authority and advisory bodies, and allows fines of up to €35 million for offending developers. Small and medium-sized enterprises face lighter requirements. Hungary transposed the EU framework into national law with Act LXXI of 2025.
For harm liability, Imre Miklós favored applying the "dangerous operation" concept: the operator who uses the system for their own interest bears responsibility for resulting damages. European Parliament recommendations suggest liability should not be excludable or limited. The idea of a joint liability insurance fund was also raised.
Imre warned that the EU’s human-centric regulatory approach is not necessarily shared by all competitors — for example, China’s different regulatory stance could create a competitive disadvantage for the EU.
Practical recommendations for AI SLAs
Koza Andrea emphasized that adopting AI is not merely a technical tool change but a full governance transformation: questions of responsibility, input data quality, and brand protection must be redesigned. She cited international court decisions: in Air Canada’s 2024 case and a 2025 case involving a German health portal, courts held that the company and its chatbot were legally identical, meaning the provider could not argue that correct information was available elsewhere.
She proposed that every AI-containing SLA rest on three pillars:
- the EU AI Regulation (EU 2024/1689),
- the ISO 42001 management standard, and
- the deployed system itself (operational parameters, oversight, data usage).
Benkóné Paulovics Anita brought an SME perspective: an EU Commission report finds Hungarian SMEs well equipped with digital tools but underutilizing them — only around 10 percent of Hungarian SMEs use AI, roughly half the EU average. Failures often stem from unclear processes during implementation; she described a company that intended to replace a well-functioning SAP system because poor implementation led staff to keep manually recording data in Excel.
She stressed the need to convince staff that AI is intended to make work more efficient rather than replace employees, and noted that AI outputs must be checked — early bots that produced incorrect VAT interpretations illustrate this point.
Conclusion
Panelists agreed that conventional SLAs alone cannot manage the novel risks introduced by autonomous AI systems. AI-specific agreements must combine legal, standards-based, and technical elements, clarify liability, and be supported by infrastructure investments that strengthen European technological sovereignty. The discussion also touched on broader ethical and societal dimensions: the Vatican and international expert bodies are engaging with AI issues, while aligning regulation and practice remains a continuing challenge.
Tags: SME, European Union, cloud, artificial intelligence, MKT, EU regulation, chatbot, AI, Pope Leo XIV, SAP
