The EU AI Act was designed to protect Europeans from the risks of artificial intelligence. Instead, critics say it is slowing the rollout of medical AI tools, driving uncertainty across the healthcare sector, and putting pressure on Europe’s medtech industry. Brussels is now retreating by delaying key provisions through the AI Omnibus package, but the changes fall far short of the regulatory reset many companies had hoped for.
The EU AI Act protects citizens from AI benefits, too
On May 7, negotiators from the Council presidency and the European Parliament reached a provisional agreement (pdf) on changes to the EU AI Act as part of the broader “Omnibus VII” simplification package, also known as the AI Omnibus. The initiative is designed to ease parts of the EU’s AI regulatory framework by delaying obligations for high-risk systems, including those used in healthcare, reducing compliance burdens, and introducing targeted simplifications for companies developing AI, especially smaller businesses.
“The agreement on the AI act significantly supports our companies by reducing recurring administrative costs,” according to Marilena Raouna, Deputy Minister for European affairs of the Republic of Cyprus.
Brussels has recognized that while the EU AI Act was created with good intentions, some provisions could effectively stifle healthcare innovation that doctors and patients are eagerly awaiting. One of the first casualties was OpenEvidence, a generative AI platform designed to support clinical decision-making. Although 42 percent of physicians in the United States reportedly use the platform, it has been unavailable in Europe since April 2026. Under the EU AI Act, such a system is classified as high-risk, imposing obligations that the company says it cannot realistically meet.
The AI Act regulates AI systems based on their risk level. The strictest requirements apply to so-called high-risk systems, including those used in healthcare. These provisions were originally scheduled to take effect in August 2026. The framework was designed to provide stronger safeguards through requirements related to transparency, documentation, and risk management. But AI developers and medtech companies warned that the rules could suppress the development of algorithms created in Europe and remove potentially valuable tools from the market. The EU AI Act was intended to protect citizens from the harmful effects of AI, but critics argue that it has also begun shielding them from its benefits.
Criticism has focused on extensive documentation requirements, vague definitions of high-risk systems, overlapping sector-specific regulations such as the AI Act and the Medical Device Regulation, the absence of technical standards, and the growing legal liability associated with rapidly evolving AI models. OpenEvidence is not the only example. Other affected technologies include ChatGPT for Clinicians, AI assistants designed to analyze medical data, and predictive algorithms. Also, AI scribes that automatically generate clinical documentation during patient visits are affected – although such tools could qualify as low-risk administrative AI, many healthcare providers prefer to avoid regulatory uncertainty and delay implementation altogether.
The AI Omnibus may disappoint innovators
The new Omnibus package eases some of these pressures. The most significant change concerns high-risk systems. Instead of August 2026, some provisions will now apply in December 2027. For AI systems already regulated as medical devices under the Medical Device Regulation, including radiology tools, diagnostic software, and critical decision support systems, the deadline has been postponed until August 2028.
The AI Omnibus also simplifies certain obligations, introduces exemptions for small and medium-sized enterprises, reduces uncertainty arising from overlapping regulations, and relaxes requirements for systems that support decision-making rather than make decisions autonomously.
Officially, the European Commission says these changes are necessary to avoid legal uncertainty and because technical standards are still not fully developed. But the broader reality is that the AI Act attempted to regulate a technology evolving so quickly that rules introduced today may become outdated tomorrow. Generative AI models are a clear example. Their outputs are often difficult to explain, and their behavior can evolve continuously through updates and adaptation. Agent-based systems are already emerging, while AI is being integrated into electronic medical record systems, blurring the line between high- and low-risk applications.
The AI Omnibus is part of the broader Digital Omnibus strategy, which aims to reduce digital bureaucracy across Europe. In healthcare, the sheer volume of regulations governing digital technologies can discourage even determined innovators. In addition to the AI Act, companies must also navigate the Medical Device Regulation, the General Data Protection Regulation, the Data Act, NIS2, and the European Health Data Space framework.
The European Commission increasingly recognizes that the AI Act may slow the development of generative AI systems for healthcare and push startups and medtech companies outside the European Union. Yet those hoping for a major liberalization are likely to be disappointed. The philosophy of the AI Act remains largely intact. The legislation still relies on risk classification, still requires transparency obligations for AI models, still creates extensive documentation requirements, and still mandates human oversight, even in situations where AI systems may outperform human operators.
European doctors are skeptical about deregulation
Proposals to loosen regulations under the Digital Omnibus have already faced criticism from the Standing Committee of European Doctors (CPME). In a position paper published in April, the organization called for a cautious approach to AI in medicine. CPME argues that some proposed changes could weaken privacy protections and make it easier to process sensitive patient data for AI development.
The organization is particularly concerned about attempts to expand the use of pseudonymized or anonymized patient data for AI training. CPME warns that weakening GDPR protections could undermine patient trust and medical confidentiality, which remain central to healthcare systems in Europe.
Supporters of deregulation argue that Europe risks falling behind the United States and China if companies continue facing excessive compliance burdens. Critics of the current framework say the continent cannot simultaneously demand AI leadership while imposing regulatory obligations that smaller developers cannot realistically meet.
The debate surrounding the AI Omnibus reflects a broader tension within European healthcare policy. Policymakers are trying to balance innovation, patient safety, privacy, and competitiveness simultaneously. Whether the EU can achieve all four goals simultaneously remains an open question.
The provisional agreement reached in May still needs formal approval from both the Council and the European Parliament. It will then undergo legal and linguistic review before the legislation can be officially adopted by the two institutions in the coming weeks.
