At the time of this updated blog post, regulatory frameworks for Artificial Intelligence (AI) in the pharmaceutical and medical device industries are in place. Since 2023, the conversation has shifted decisively from "what should the rules for AI be?" to "how do we implement AI?", and 2025 became the year AI moved from exploration to real-world regulatory use.
The use of AI has made rapid progress and is already benefiting the healthcare industry and patients. In June 2025, the first generative-AI-designed drug (where AI proposed both the biological target and the molecule) reported positive Phase IIa clinical results. This milestone would have seemed distant only two years earlier, yet these applications still hold far more potential for the future.
To ensure the safe and ethical implementation of AI, governments and regulators have built, and are now operating, a comprehensive legal framework that strikes a balance between potential risks and benefits.
In Europe, the European Medicines Agency (EMA) and the Heads of Medicines Agencies (HMAs) are executing a multi-annual AI workplan for 2023-2028. EMA finalised its Reflection Paper on the use of AI across the medicinal product lifecycle in September 2024, and the EU Artificial Intelligence Act (EU AI Act), the world's first comprehensive AI law. It entered into force on 1 August 2024 and is now in its phased application period.
In the United States, the FDA issued its first dedicated draft guidance on AI in drug and biological product development in January 2025 and is deploying generative AI internally across the agency.
Globally, the World Health Organization (WHO), the International Coalition of Medicines Regulatory Authorities (ICMRA), the OECD, the UK's MHRA, the Council of Europe and the G7 have all advanced AI recommendations, principles and binding instruments.
In a notable sign of convergence, EMA and the FDA published a joint set of "Guiding Principles of Good AI Practice in Drug Development" in January 2026.
With a particular focus on healthcare and medical devices, these developments reflect a maturing recognition that clear, operational regulatory frameworks are necessary to ensure the safe and ethical use of AI.
AI in Pharma
When we talk about AI in pharma, we typically refer to a technique, model or algorithm integrated into computer systems that enables AI to learn and reason with data. This model allows it to perform automated tasks and make decisions and predictions without explicit programming of every step by a human.
This ability is behind phenomena such as Machine Learning (ML), Natural Language Processing (NLP), computer vision, conversational intelligence, and neural networks.
Since 2023, the centre of gravity has shifted towards generative AI and large language models (LLMs), and increasingly towards large multi-modal models (LMMs) that combine text, images and other data types, as well as early "agentic" AI systems that can carry out multi-step tasks with limited supervision.
ML refers to machines learning from data and improving their performance over time. NLP refers to the ability of computers to understand text and spoken words in much the way humans can.
As a result of Deep Learning (DL), a subset of ML, AI systems can identify patterns, predict outcomes and adapt their behaviour to changing conditions.
With advances in infrastructure such as the cloud, in hardware and software technologies, in DL models, and with the growth of big data, more information is available than ever before in the life sciences. With its advanced algorithms, AI is rapidly transforming the healthcare industry.
Every regulatory agency increasingly recognises the pivotal role AI plays in the life sciences. It is reshaping public health, clinical practice, research, drug development, disease surveillance and industry management, driving innovation and enhancing patient outcomes.
Despite the enthusiasm, AI in pharma remains controversial regarding bias, privacy and safety. Also, generative AI has added new concerns such as "hallucinations" (confident but fabricated outputs), traceability of AI-generated content, and data provenance.
In a study titled Artificial intelligence in healthcare, the Scientific Foresight Unit ("STOA") of the European Parliamentary Research Service analysed AI applications, risks and opportunities, and ethical and social impacts.
The study highlighted that AI in pharma can automate repetitive tasks and help doctors to diagnose and treat illness, while also carrying risks: AI errors, misuse of biomedical AI tools, AI biases, lack of transparency, data privacy and security issues, gaps in accountability and obstacles to implementation.
For example, biometrics and facial recognition technologies require complex AI algorithms to process vast amounts of information. This raises concerns about privacy and fundamental rights. All in all, there are many ethical questions to consider.
The design, development and deployment of AI technologies for health must therefore account for ethical considerations and human rights.
As a result, it remains essential to develop regulatory oversight mechanisms that hold organisations accountable and responsive to those who may benefit from AI products and services, and that ensure transparency.
The principle now embedded across EMA, FDA, WHO and OECD texts is a risk-based, human-centric approach: the higher an AI tool's influence on a regulatory or clinical decision, and the greater the consequence of that decision, the more rigorous the evidence, oversight and lifecycle monitoring required.
Potential AI applications in the life sciences industry
AI is transforming the life sciences industry in many ways, from drug discovery and development to personalised medicine and disease diagnosis.
Market estimates vary widely depending on whether they measure AI in life sciences, AI in pharma or AI in drug discovery, but all point to rapid double-digit growth.
For drug discovery specifically, Grand View Research (2025) values the market at around USD 2.3 billion in 2025, rising to roughly USD 13.8 billion by 2033 (about 24.8% Compound annual growth rate, CAGR).
For the broader AI-in-life-sciences market, MarketsandMarkets (2026) estimates around USD 21.6 billion in 2026, growing to roughly USD 69 billion by 2031 (about 26% CAGR).
Investment has accelerated sharply: by 2025, AI partnering had become standard among large pharma, with multibillion-dollar collaborations such as Eli Lilly-Insilico Medicine, AstraZeneca-CSPC and Sanofi-Helixon. This underscores the significant interest and investment in AI across the sector.
AI has the potential to improve efficiency, speed, accuracy and patient outcomes.
The life sciences industry has many potential applications for AI, including:
- Drug discovery
- Personalized medicine
- Medical imaging
- Clinical trial design
- Digital twins
- Real-world evidence (RWE) generation at scale
- Medical writing and regulatory dossier drafting
- Pharmacovigilance and real-time safety-signal detection
- Demand forecasting
- Smart advertising
- Intelligent supplier sourcing
- Manufacturing optimization
- Support for SOPs
As a concrete sign of clinical traction, the number of AI-originated drug programmes in clinical development rose from roughly two dozen in late 2023 to well over a hundred by early 2026. No AI-discovered drug has yet been approved, and there have been setbacks as well as successes, but the pipeline is maturing quickly.
Current situation and regulations
AI regulations and policies have moved well beyond the drawing board. Several frameworks are now in force or in phased application. Let's review the main milestones and their status as of mid-2026.
Worldwide
In May 2019, the Organisation for Economic Co-operation and Development (OECD) adopted the OECD Principles for AI (the Recommendation), the first intergovernmental AI standard. Its objectives include promoting innovation and trust in AI through responsible stewardship of trustworthy AI while respecting human rights and democratic values.
The Principles were significantly updated in two stages: in November 2023 the definition of an "AI system" was revised (notably to cover generative AI and to align with the EU AI Act), and in May 2024 the Principles themselves were strengthened, adding provisions on misinformation and information integrity, misuse, safe decommissioning of systems, and environmental sustainability.
In June 2021, WHO published guidance on the Ethics and Governance of Artificial Intelligence for Health, outlining six consensus principles to ensure AI benefits humankind and identifying ethical challenges and risks. This introduced foundational knowledge for policymakers, AI developers and healthcare providers.
In October 2023, WHO published Regulatory considerations on artificial intelligence for health, exploring regulatory and health-technology-assessment concepts and good practices. There were 18 considerations across areas such as transparency, risk management, validation, data quality and privacy.
In March 2025, WHO issued published Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models, with more than 40 recommendations on generative AI across diagnosis and clinical care, patient self-use, administrative tasks, education, and scientific research and drug development. This is now WHO's principal generative-AI guidance for health.
Two binding and semi-binding international instruments have also emerged. The Council of Europe Framework Convention on Artificial Intelligence, the first legally binding international AI treaty, was adopted in May 2024 and opened for signature in September 2024. The EU ratified it in May 2026, although as of mid-2026 the treaty has not yet entered into force (it requires five ratifications).
The G7 Hiroshima AI Process (international guiding principles and a code of conduct, 2023) gained an operational, OECD-hosted reporting framework in February 2025.
These developments, together with the European and US work described below, continue to shape the future of AI across the medicines lifecycle.
EMA in Europe
The EMA is ensuring that they stay up to date with AI advances to remain at the forefront of the industry.
Here are the key moments in Europe:
In April 2021, the European Commission published the first proposal for the Artificial Intelligence Act ("EU AI Act"), the first attempt at harmonised rules on AI.
In December 2023, EMA and the HMAs published a multi-annual AI workplan 2023-2028 to build a regulatory system that maximises the benefits of AI while mitigating risk. Governance of this workplan has since moved from the former Big Data Steering Group to the new Network Data Steering Group (NDSG), which now carries the work forward under an expanded 2025-2028 plan.
To monitor progress, EMA established an AI Observatory, which published its first annual report in 2025 and its second in June 2026; the latter characterised 2025 as the year European regulators moved "from AI exploration to real-world implementation."
A landmark of that implementation came in March 2025, when EMA's CHMP issued its first qualification opinion on an AI tool, known as AIM-NASH, which helps pathologists score liver-biopsy severity in clinical trials for steatohepatitis. It was the first time EMA accepted data generated with a (human-supervised) AI tool as scientifically valid for regulatory purposes.
In September 2024, the Big Data Steering Group of the EMA published “Reflection paper on the use of Artificial Intelligence (AI) in the medicinal product lifecycle”. It sets out a risk-based approach across the full lifecycle (drug discovery, non-clinical, clinical trials, manufacturing and pharmacovigilance), emphasising human oversight, data integrity and continuous performance monitoring. The paper also stresses that AI use must comply with existing law, consider ethics and respect fundamental rights.
On the legislative side, the EU AI Act moved from political agreement to law. It entered into force on 1 August 2024 and applies in phases:
- Prohibited "unacceptable-risk" practices and AI-literacy obligations from February 2025.
- Obligations for general-purpose AI (GPAI) models from August 2025,
- GPAI Code of Practice published in July 2025.
- Transparency obligations from August 2026.
The Act is risk-based: most medical device AI is classified as "high-risk" because it falls under the Medical Devices and IVD Regulations.
Importantly, AI used purely in drug discovery, clinical analysis or manufacturing is generally not automatically high-risk under the AI Act (medicines are not listed among the Act's product legislation); such pharma AI is governed primarily by EMA guidance and the EU pharmaceutical legislation.
In a significant late-stage development, a Digital Omnibus simplification package agreed politically in May 2026 is set to postpone the high-risk obligations to December 2027 for stand-alone high-risk systems, and to August 2028 for AI embedded in regulated products such as medical devices. These deferred dates become legally binding only once formally published.
Finally, the EU's broader pharmaceutical legislation reform, the largest overhaul in two decades, was adopted in March 2026. The pharma package introduces regulatory sandboxes into EU pharma law for the first time and anticipates AI-enabled, more adaptive post-market safety monitoring, with phased application later this decade.
Source: Downloaded whitepaper
Let’s now have a look at AI’s development work in the US.
FDA in the US
The FDA is committed to ensuring drugs are safe and effective while facilitating innovation. It has engaged extensively with stakeholders to understand the implications of AI-based technologies for drugs and medical devices, and has now begun issuing dedicated AI guidance.
Let’s discuss what is the current situation:
In January 2019, the FDA published Developing a Software Pre-certification Program: A Working Model, aimed at more streamlined oversight of software-based medical
devices.
In April 2019, the FDA published a discussion paper, Proposed Regulatory Framework for Modifications to AI/ML-Based Software as a Medical Device (SaMD), describing its approach to premarket review for AI/ML-driven software modifications.
In January 2021, the FDA published the AI/ML-Based Software as a Medical Device (SaMD) Action Plan, outlining five goals:
- A tailored regulatory framework for AI/ML-based SaMD
- Good Machine Learning Practice (GMLP)
- A patient-centred approach incorporating transparency for users
- Regulatory science methods related to algorithm bias & robustness
- Real-world performance (RWP) monitoring
The FDA released two discussion papers in 2023 (on AI/ML in drug and biological product development, and on AI in drug manufacturing) to stimulate debate among industry, academia, patients and regulators. These have since been built upon by formal guidance:
- The first paper, Artificial Intelligence and Machine Learning in the Development of Drug & Biological Products aims to stimulate discussion between interested parties such as pharmaceutical companies, ethicists, academia, patients, and global regulatory and other authorities. Also, on the potential use of AI/ML in drug and biological development, and the development of medical devices to use with these treatments. The article also discusses ways to address possible concerns and risks associated with AI/ML.
- The second paper Artificial Intelligence in Drug Manufacturing from the FDA seeks public feedback on regulatory requirements applicable to the approval of drugs manufactured using AI technologies.
The most significant recent steps are:
- Drug and biological products (January 2025): the FDA issued its first dedicated draft guidance, "Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products". It proposes a risk-based credibility-assessment framework built around the model's "context of use" and "model risk," with a seven-step plan to establish how much trust can be placed in an AI model's output. As of mid-2026 it remains a draft, with finalisation anticipated.
- Medical devices (December 2024): the FDA finalised its guidance on Predetermined Change Control Plans (PCCPs) for AI-enabled device software functions, allowing manufacturers to pre-specify and obtain authorisation for future model changes without a new submission each time.
- Medical devices (January 2025): the FDA issued draft guidance on AI-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations, taking a total-product-lifecycle approach; still in draft as of mid-2026.
The cumulative effect is visible in the numbers: the FDA's public list of AI-enabled medical devices has grown past roughly 1,000 authorisations (the large majority in radiology), and CDER reports having received several hundred drug and biologic submissions containing AI components.
The FDA is also adopting AI internally. In June 2025 it launched "Elsa," an agency-wide generative-AI tool to help staff summarise documents, compare labels and prioritise inspections (operating in a secure environment that does not train on regulated-industry data). In May 2026 it expanded this with "Elsa 4.0" and a consolidated internal data platform.
The US policy environment also shifted markedly in 2025. A January 2025 executive order, "Removing Barriers to American Leadership in Artificial Intelligence," set a deregulatory, pro-innovation tone, and a major HHS restructuring affected FDA staffing, including some digital-health and AI reviewers. The interagency picture is coordinated through bodies such as the CDER AI Council, and in November 2024 the FDA's Digital Health Advisory Committee held its first meeting, focused on generative-AI-enabled devices.
In a clear signal of international alignment, the FDA and EMA jointly published "Guiding Principles of Good AI Practice in Drug Development" in January 2026. The publication contained a non-binding set of ten principles intended to underpin future AI-specific guidance on both sides of the Atlantic.
Other regulators and international bodies
Beyond the EU and US, several regulators have advanced quickly.
The UK's MHRA launched the "AI Airlock," the world's first regulatory sandbox for AI as a medical device, in 2024, ran successive cohorts through 2025-2026, and secured multi-year funding for a third phase.
The UK also established a National Commission into the Regulation of AI in Healthcare in late 2025. ICMRA's 2021 horizon-scanning report on AI remains its core reference, with substantive follow-up having migrated to EMA.
At the international standards level, IMDRF finalised Good Machine Learning Practice guiding principles for medical device development in January 2025.
And ICH finalised ICH M15 guidance on model-informed drug development, which explicitly covers AI/ML models, in January 2026.
National regulators in China (NMPA) and Japan (PMDA) have also expanded AI-device review pathways and change-control mechanisms.
Artificial Intelligence Workplan
In December 2023, EMA and the HMAs developed a multi-annual AI workplan 2023-2028 within the European medicines regulatory network (EMRN).
Its governance now sits with the Network Data Steering Group (NDSG), which has carried the work into an expanded 2025-2028 plan.
The plan promotes the responsible use of AI across the pharmaceutical and medical device industries. Its initiatives include implementing and monitoring AI for internal regulatory purposes, enhancing network-wide analytics capability, and collaborating with international and EU agencies.
To protect patient safety, the EU is committed to regulating AI effectively and responsibly, emphasising stakeholder communication, research priorities and guiding principles. This reflects the EU's commitment to the ethical and responsible use of AI in the life sciences.
Through 2028, the workplan addresses four key dimension:
1. Guidance, policy and product support: Preparing the network for the EU AI Act, and operating an AI Observatory to monitor the impact of AI and the emergence of new systems and approaches. The Observatory has now published annual reports in 2025 and 2026, and EMA finalised its AI reflection paper (September 2024) and issued LLM guiding principles for staff.
2. Tools and technologies: Rolling out knowledge-mining tools across the network, piloting and monitoring large language models and related chatbots, surveying the network's data-analysis capability, and publishing a network tools policy for open, collaborative AI development. EMA's AI-enabled "Scientific Explorer" knowledge-mining tool, introduced in 2024, was extended in 2026 to cover marketing-authorisation applications and assessment reports.
3. Collaboration and change management: Working with EU agencies, international partners, an AI virtual community and experts in AI, medical devices and academia. The clearest output is the EMA-FDA joint guiding principles of good AI practice in drug development (January 2026).
4. Experimentation. Running experimentation cycles, developing internal guiding principles for responsible AI, issuing technical deep dives in areas such as digital twins, and publishing and periodically revising a roadmap of network research priorities.
The future of AI in the life sciences: what to expect
In the near future we can expect existing frameworks to be finalised and operationalised rather than merely proposed:
- The FDA's drug-AI guidance moving from draft to final.
- The EU AI Act's high-risk obligations taking effect (on the timeline confirmed by the Digital Omnibus).
- EMA building further qualification opinions and guidance on the foundation of its 2024 reflection paper.
AI must be regulated effectively to ensure its ethical and responsible use and to protect patient safety.
The rise of generative and agentic AI sharpens long-standing challenges, the difficulty of understanding complex models, ensuring transparency in decision-making, and managing risks such as hallucinations, bias and data provenance.
For AI to succeed and benefit society, governments, organisations and experts must collaborate.
The growing convergence between EMA, the FDA and international bodies, expressed in shared principles and aligned, risk-based approaches, suggests this is happening.
Experts continue to recommend a holistic approach, addressing intended use, continuous learning, human oversight and the quality of training data, while the community works towards shared understanding and mutual learning.
Discover how QA is evolving into a data-literate, digitally fluent function, guiding AI-powered systems and teams to protect trust as complexity accelerates.
Final thoughts
The integration of AI holds both promise and challenges in the life sciences. As regulatory frameworks move from proposal to application and industry leaders navigate ethical questions, a transformative journey is well under way.
The key takeaways are:
Discover how to surf the wave of AI innovation with Scilife smart QMS.
