Generative Era of Medical AI: Landscape of Agents, Biology and Governance

14

1. Introduction: The Year of Agentic Scale

The trajectory of artificial intelligence in medicine has historically been defined by cycles of exuberant promise followed by the sobering reality of implementation hurdles. However, the year 2025 marks a definitive departure from this pattern, signaling the commencement of the “Generative Era” in healthcare. This period is characterized not merely by the refinement of algorithmic precision but by the fundamental operationalization of generative artificial intelligence (AI) across the entire continuum of care—from the molecular design of novel therapeutics to the administrative architecture of hospital systems. The narrative of 2025 is no longer one of “potential” or “pilots”; it is a narrative of scale, agentic autonomy and the rigorous quantification of value.

If 2023 was the year of the chatbot and 2024 the year of the pilot, 2025 has firmly established itself as the year of the “Agent.” The industry has moved beyond passive interfaces that simply respond to queries, embracing “Agentic AI” systems capable of autonomous reasoning, multi-step planning and execution within complex digital environments.¹ These agents do not merely summarize information; they navigate electronic health records (EHRs), formulate research hypotheses, design experimental protocols and draft regulatory submissions. The shift is profound: AI is transitioning from a tool used by humans to a collaborator working alongside humans.

This report provides an exhaustive analysis of the state of medical AI in 2025. It synthesizes data from industry-leading reports, peer-reviewed literature from journals such as Nature Medicine and NEJM AI and regulatory guidance from the FDA and European Commission. The analysis reveals a sector that is charging ahead with unprecedented velocity—where 94% of healthcare organizations are actively using or piloting AI and where the integration of “genomic foundation models” is beginning to decode the fundamental language of life itself.²

Yet, this transformation is not devoid of friction. The landscape is bifurcated between “scalers”—organizations that have successfully moved into enterprise-wide production—and those still trapped in “pilot purgatory”.¹ Furthermore, the rapid capabilities of these systems have necessitated a complete overhaul of regulatory frameworks, resulting in a complex new tapestry of governance that spans the United States and the European Union.⁵ This report dissects these developments, offering a granular view of the technological, clinical and operational realities of healthcare in 2025.

2. The Macro-Economic and Operational Landscape

The economic and operational indicators of 2025 paint a picture of a healthcare sector that has committed definitively to an AI-driven future. The hesitancy that characterized previous years has largely evaporated, replaced by a strategic imperative to deploy generative capabilities to ensure financial viability and clinical excellence.

2.1 From Experimentation to Enterprise Value

The maturation of the market is evident in the depth of adoption. According to NVIDIA’s “State of AI in Healthcare and Life Sciences: 2025 Trends” report, the adoption rate among healthcare professionals has reached a staggering saturation point. Approximately 63% of surveyed professionals report active use of AI, with an additional 31% currently piloting initiatives.² This combined engagement rate of 94% positions healthcare as a global leader in AI adoption, surpassing other major industries which average only 50% uptake.²

However, adoption is not uniform. McKinsey’s 2025 “State of AI” report introduces a critical nuance: while experimentation is rampant, scaling remains the primary challenge. Nearly two-thirds of organizations report they have not yet successfully scaled AI across the enterprise.¹ This distinction creates a “maturity gap” where a select group of leading organizations—the “scalers”—are reaping the majority of the benefits, while the “experimenters” continue to grapple with infrastructure and governance hurdles.

Despite these challenges, the leading indicators for value generation are overwhelmingly positive. A critical finding in 2025 is the rapid realization of Return on Investment (ROI). Nearly half of the organizations deploying AI report seeing a positive ROI within a single year of deployment.² This accelerated payback period is driving aggressive budget expansions, with 78% of organizations planning to increase their AI expenditures in the coming fiscal year.²

The nature of this value is dual-pronged:

1. Revenue Generation: Contrary to the early narrative focused solely on cost-cutting, 81% of respondents in the NVIDIA study reported that AI contributed to increased revenue.² In the pharmaceutical sector, this is linked to faster R&D cycles (cited by 41% of firms), allowing drugs to reach the market sooner. In provider settings, revenue gains are driven by optimized patient throughput, improved coding accuracy and the reduction of claim denials.
2. Operational Efficiency: Cost reduction remains a staple benefit, with 73% of organizations reporting reduced operational costs.² This is achieved largely through the automation of administrative tasks, which liberates human capital for higher-value clinical work.

Table 1: Key Economic Indicators of Medical AI Adoption (2025)

Metric	Value	Implications	Source
Total Engagement	94% (63% Active, 31% Piloting)	AI has moved from niche to normative; non-adoption is now a competitive disadvantage.	2
Revenue Impact	81% report increased revenue	Shifts the narrative from “cost-saving” to “growth-driving.”	2
Operational Impact	73% report reduced costs	Validates the efficiency thesis of generative automation.	2
ROI Timeline	~50% within 1 year	remarkably short payback period encourages rapid reinvestment.	2
Budget Outlook	78% increasing budgets	Signals a long-term capital commitment to AI infrastructure.	2

2.2 The Rise of Agentic AI

The most significant technological shift observed in 2025 reports is the graduation from “Generative AI” (systems that create content) to “Agentic AI” (systems that execute tasks). McKinsey reports that 62% of organizations are now experimenting with AI agents.¹

The distinction is critical. A generative chatbot might draft a prior authorization letter. An AI Agent, however, acts autonomously: it identifies that a prior authorization is needed, retrieves the necessary clinical evidence from the patient’s history, navigates the payer’s portal, submits the request and monitors the status—all with minimal human intervention. This agentic capability addresses the “last mile” problem in healthcare automation, where the friction of switching between applications often negated the time saved by the AI.

This agentic evolution is driven by the demand for deeper integration into complex workflows. In 2025, LinkedIn hiring trends indicate a sharp rise in demand for “Agentic AI Developers,” reflecting the industry’s pivot toward building systems that can reason and act.⁷ The U.S. Food and Drug Administration (FDA) itself has adopted this technology, deploying agentic AI capabilities within a high-security GovCloud environment to streamline its own regulatory review processes.⁸ This endorsement by a regulator underscores the maturity and reliability of agentic architectures.

2.3 Strategic Deployments and the “Flywheel Effect”

The year 2025 has witnessed deployments of a scale previously unimagined. The era of the “pilot” in a single department is over; the era of system-wide rollout has begun.

Kaiser Permanente’s Historic Rollout:

In what is described as the largest generative AI rollout in healthcare history, Kaiser Permanente deployed Abridge’s ambient documentation solution across 40 hospitals and more than 600 medical offices.4 This implementation was characterized as Kaiser’s fastest technology adoption in over 20 years. The speed of this rollout speaks to the acute pressure on health systems to alleviate workforce burnout. By moving from a pilot to a system-wide standard of care, Kaiser is creating a data “flywheel”: the massive volume of interactions processed by the AI refines the model’s accuracy specifically for Kaiser’s diverse patient population, creating a self-reinforcing loop of quality improvement.

Advocate Health’s rigorous Selection:

Advocate Health’s approach illustrates the rigor now applied to AI procurement. The system evaluated over 225 AI solutions to select just 40 use cases for live deployment.4 This 18% selection rate indicates a high bar for entry. Their deployments include the largest implementation of Microsoft Dragon Copilot, alongside imaging tools like Aidoc and Rad AI. This strategic filtration ensures that only tools with proven clinical or operational utility are allowed to touch the enterprise workflow.

Market Growth in Business Intelligence:

The supportive infrastructure for these deployments is also booming. The U.S. Business Intelligence (BI) software market, now heavily integrated with generative AI for analytics, is projected to grow from $11.35 billion in 2024 to $27.49 billion by 2033.9 This growth is driven by the need to visualize and interpret the massive data streams generated by these new AI agents.

3. The Biological Foundation: Generative Biology and the “Design Room”

While operational AI transforms the hospital, a quieter but perhaps more profound revolution is occurring in the laboratory. 2025 is the year generative AI mastered the “language of life.” Researchers have successfully mapped the principles of Large Language Models (LLMs) to biological sequences—DNA, RNA and proteins—creating “Genomic Foundation Models” that can read, write and edit biological code with superhuman proficiency.

3.1 Evo 2: The Genomic Generalist

A landmark achievement of 2025 is the release of Evo 2, a biological foundation model developed by a coalition of researchers from the Arc Institute, Stanford, UCSF and UC Berkeley.³ Unlike previous models that were limited to specific domains (like human text or single viral genomes), Evo 2 is a “generalist” trained on the DNA of over 100,000 species across the entire tree of life, including bacteria, archaea and eukaryotes (humans and plants).³

Architectural Breakthroughs:

Evo 2 operates on a scale comparable to the most powerful linguistic models, boasting 7 billion parameters.11 Its most critical technical innovation is its context window of 131 kilobases (131,000 nucleotides) at single-nucleotide resolution.11

• Why this matters: In biology, a gene at one end of a chromosome often interacts with a regulator millions of base pairs away. Traditional models with short context windows miss these long-range interactions. Evo 2’s massive context allows it to “see” the entire genomic landscape, understanding the complex dependencies that govern life.

Capabilities:

• Zero-Shot Prediction: Evo 2 can predict the functional impact of genetic mutations without specific training on that disease, identifying disease-causing variants in human genes with high accuracy.³
• Generative Design: The model can generate novel genomes as long as those of simple bacteria. This capability to “write” viable DNA sequences de novo represents a monumental leap in synthetic biology.¹⁰
• Gene Editing Discovery: It aids in identifying new components for CRISPR systems, effectively expanding the toolkit for genetic engineering.¹²

3.2 AlphaGenome: Decoding the “Dark Matter”

Complementing the broad scope of Evo 2 is Google DeepMind’s AlphaGenome, released in mid-2025.¹³ While Evo 2 focuses on the breadth of the tree of life, AlphaGenome focuses on the depth of regulation.

The human genome is largely composed of non-coding DNA—regions that do not make proteins but regulate when and where genes are active. This “dark matter” harbors the majority of genetic variants associated with complex diseases (like Alzheimer’s or diabetes), yet its rules have remained opaque. AlphaGenome accepts input sequences of up to 1 million base pairs and predicts thousands of molecular properties, such as epigenetic marks, RNA splicing sites and gene expression levels.¹³

By allowing researchers to virtually mutate a single letter in a non-coding region and observe the ripple effects on gene regulation, AlphaGenome provides a causal bridge between genetic association and biological mechanism.¹³ It effectively serves as a “virtual cell,” allowing scientists to run millions of experiments in silico to prioritize targets for drug discovery.

3.3 Generative Protein Design: Outperforming Nature

In the realm of proteins—the molecular machines that perform life’s functions—2025 has proven that AI can not only mimic nature but surpass it.

AlphaFold 3 and the Interaction Universe:

Building on the Nobel Prize-winning success of AlphaFold 2, the 2025 release of AlphaFold 3 has redefined structural biology.14 While AlphaFold 2 predicted the shapes of single proteins, AlphaFold 3 predicts how proteins interact with everything else: DNA, RNA, small molecule drugs (ligands) and ions.16

Using a “diffusion” architecture (similar to the technology behind image generators like DALL-E), AlphaFold 3 builds molecular structures from noise, achieving unprecedented accuracy in predicting how drugs bind to their targets.16 However, independent benchmarks in 2025 have tempered the hype slightly, noting that while it excels at static interactions, it still struggles with dynamic conformational changes (large movements) and shows some bias toward active states in certain receptor types.17

The “Super-Natural” TNIK Inhibitor:

A definitive proof-of-concept for AI drug design was published in Nature Biotechnology in 2025. Insilico Medicine utilized generative AI to discover and design a novel inhibitor for TNIK, a cancer target.18 Crucially, the AI-designed molecule (INS018_055) and a novel PROTAC degrader (D16-M1P2) demonstrated properties superior to traditional inhibitors, including dual-action mechanisms and higher selectivity.19 This study was notable for disclosing the raw experimental data, providing the industry with a transparent roadmap of the AI discovery process.18

Synthetic Biology and the “Design Room”:

This capability to design novel biological entities has created what researchers call a “Design Room”—a theoretical space where AI can explore biological possibilities that evolution never reached.20 For instance, AI is being used to design “NGT-1” plants (New Genomic Techniques) with specific traits like insect resistance. This raises complex regulatory questions: if an AI designs a gene sequence that could theoretically exist in nature but doesn’t, is it a GMO? 2025 regulatory frameworks are just beginning to grapple with these “designed” organisms.20

4. Clinical Intelligence and Diagnostic Reasoning

Moving from the lab to the clinic, 2025 has seen generative AI deployed as a high-level diagnostic partner. The debate has shifted from “can AI pass the exam?” to “can AI diagnose the patient?”

4.1 Med-Gemini: The Multimodal Standard

Google’s Med-Gemini, a family of models fine-tuned for medicine, represents the state-of-the-art in 2025. While it achieved a record-breaking 91.1% on the MedQA benchmark (USMLE-style questions), its true value lies in its multimodal capabilities.²¹

Medicine is rarely text-only; it is a synthesis of history (text), radiology (images) and pathology (slides). Med-Gemini excels at this synthesis.

• Needle-in-a-Haystack Retrieval: In long, complex electronic health records (EHRs), finding a specific detail (e.g., “did the patient have a reaction to lisinopril in 2018?”) is challenging. Med-Gemini utilizes long-context processing to retrieve these “needles” with high precision.²³
• 3D Volume Understanding: Unlike previous models that looked at single 2D images, Med-Gemini can process 3D CT volumes and surgical videos, allowing it to answer questions about anatomical relationships and surgical phases.²¹
• Radiology Reporting: In head-to-head comparisons, Med-Gemini’s generated chest X-ray reports were rated as “equivalent or better” than radiologist reports in over 50% of cases, signaling a future where the AI drafts the report for human review.²⁴

4.2 The Reality Gap: Benchmarks vs. Bedside

Despite these technical triumphs, 2025 literature highlights a persistent gap between standardized testing and real-world performance. A systematic review of Large Language Model (LLM) accuracy published in JAMA and NEJM AI reveals that clinical reasoning in the wild is far harder than on a test.²⁵

The Complexity Penalty:

While advanced models like Claude 3.7 and GPT-4 achieve >90% accuracy on “common” clinical scenarios, their performance drops significantly on “complex” cases derived from journals like the New England Journal of Medicine. On these challenging cases, accuracy hovered around 45-48%.26 While this often exceeded the performance of unassisted physicians (who averaged ~20% on these ultra-difficult cases), it demonstrates that AI is not yet infallible.

The Conversational Cliff:

A critical study in medRxiv exposed a weakness in “agentic” diagnosis. When models were tested on static vignettes (where all information is presented at once), they performed well (~80% accuracy). However, when forced to engage in a “conversation” to ask the patient questions and gather the history iteratively, performance plummeted to ~30-50%.27 This suggests that while AI is excellent at synthesizing data, it still struggles with the art of investigating—knowing what question to ask next.

Cognitive Bias:

Furthermore, AI models have been shown to inherit human cognitive biases. For example, the “framing effect” was observed in AI recommendations: models were more likely to recommend surgery for lung cancer if the outcomes were presented as “survival rates” rather than “mortality rates,” mirroring a known irrational bias in human psychology.28

4.3 Generative Radiology: Beyond Diagnosis to Reconstruction

In radiology, generative AI is playing a role beyond just “spotting the tumor.” Generative Adversarial Networks (GANs) and diffusion models are being used for image reconstruction and enhancement.

• Denoising and Safety: Generative models are used to “denoise” low-dose CT scans, effectively allowing high-quality imaging with a fraction of the radiation dose to the patient.²⁹
• Synthetic Data Generation: To train AI on rare diseases where data is scarce, researchers are using generative AI to create synthetic MRI and CT images. These images are statistically identical to real scans but contain no private patient data, solving privacy bottlenecks.³⁰
• Modality Translation: AI is now capable of “translating” an MRI image into a synthetic CT image (or vice versa), which is particularly useful for radiation therapy planning where CTs are required but MRIs provide better soft-tissue contrast.²⁹

5. The Operational Revolution: Ambient Intelligence

While the diagnostic applications of AI are intellectually captivating, the most immediate “quality of life” improvements in 2025 have come from Ambient Intelligence. This technology is aggressively attacking the administrative burden that has plagued healthcare for decades.

5.1 The Death of the Keyboard: Ambient Scribes

The adoption of ambient AI scribes—technologies that listen to the patient-encounter and autonomously draft the clinical note—has moved from a “nice-to-have” to a retention strategy for health systems.

Evidence from the Field:

A multicenter prospective study involving 263 physicians across 6 health systems provided robust data on the impact of this technology in 2025. The results were statistically significant:

• Burnout Reduction: Clinician burnout rates dropped from 51.9% pre-intervention to 38.8% after just 30 days of using the AI scribe.³²
• Cognitive Load: Physicians reported a measurable decrease in “cognitive task load,” allowing them to focus more mental energy on the patient rather than the computer.³²
• “Pajama Time”: There was a significant reduction in time spent documenting after hours, directly addressing the work-life balance crisis in medicine.³³

This technology has paradoxically used AI to make medicine “more human.” By removing the physical barrier of the computer screen, physicians report being able to make eye contact and engage in active listening, secure in the knowledge that the AI is capturing the details.³⁴

5.2 Patient-Facing AI: Clarity and Access

Generative AI is also being turned outward to face the patient.

• Discharge Summaries: One of the most dangerous times in a patient’s journey is the transition home. Standard discharge summaries are often filled with impenetrable medical jargon. In 2025, Randomized Controlled Trials (RCTs) are underway and reporting early data on AI-generated “patient-friendly” discharge instructions.³⁵ These summaries translate complex clinical concepts into plain language (e.g., converting “ambulate ad lib” to “walk as much as you feel able”), improving patient comprehension and potentially reducing readmissions.³⁷
• Navigation Assistants: Payers and retailers like CVS Health have deployed generative assistants to help members navigate benefits. These agents can explain complex coverage details (“Is my physical therapy covered?”) in natural language, reducing friction in accessing care.³⁸

6. The Regulatory and Ethical Tapestry

The explosion of capability in 2025 has forced regulators to move at an unprecedented speed. The governance landscape has shifted from vague principles to concrete, enforceable requirements.

6.1 United States: The FDA and the PCCP

The FDA has solidified its framework for adaptive AI through the Predetermined Change Control Plan (PCCP).

• The Problem: Historically, if an AI model “learned” and changed, it needed a new FDA approval, which froze models in time.
• The 2025 Solution: Finalized guidance in 2025 allows manufacturers to submit a PCCP.⁵ This plan details how the model will be retrained, what data will be used and what performance metrics must be met. If the model evolves within these pre-approved guardrails, no new submission is needed. This is a regulatory innovation that matches the iterative nature of generative AI.
• Lifecycle Management: New draft guidance on “AI-Enabled Device Software Functions Lifecycle Management” places a heavy emphasis on post-market monitoring. Manufacturers must now have continuous surveillance systems to detect “model drift” or performance degradation in real-world settings.⁴⁰

6.2 European Union: The AI Act and MDR Complexity

In Europe, the regulatory environment is more stringent. The full application of the EU AI Act in 2025 has created a complex interplay with the existing Medical Device Regulation (MDR).

• High-Risk Classification: Most medical AI is classified as “High-Risk” under the AI Act. This triggers requirements for fundamental rights impact assessments, rigorous data governance (to prevent bias) and transparency logs.⁴¹
• The Overlap Problem: A medical device software (MDSW) is now subject to both MDR and the AI Act. To address confusion, the Medical Device Coordination Group (MDCG) released guidance MDCG 2025-6 in June 2025.⁴² This document clarifies the roles: a “manufacturer” under MDR is a “provider” under the AI Act, but the “user” (doctor) is a “deployer.” This distinction places new legal burdens on hospitals (“deployers”) to ensure they are using AI in accordance with its instructions and monitoring for issues.⁴²

6.3 Global Governance: WHO and NAM

• World Health Organization (WHO): In 2025, the WHO released seminal guidance on Large Multi-Modal Models (LMMs).⁴³ This document outlines over 40 recommendations, treating LMMs not just as software but as socio-technical systems. It warns of “automation bias” (trusting the AI too much) and emphasizes the need for equitable access so that AI does not widen the global health gap.
• National Academy of Medicine (NAM): The NAM published “Generative Artificial Intelligence in Health and Medicine: Opportunities and Responsibilities for Transformative Innovation”.⁴⁵ This special publication serves as a code of conduct for the US, urging health systems to prioritize “human-in-the-loop” governance and to view AI integration as a systemic transformation rather than a tech upgrade.

7. Institutional Blueprints: Case Studies in Transformation

How are these technologies and regulations coalescing on the ground? Two major institutions illustrate the divergent yet complementary strategies of 2025.

7.1 Mayo Clinic: The Platform Strategy

Mayo Clinic has adopted a “Platform” strategy, positioning itself as a hub for AI development.

• Supercomputing: In a strategic partnership, Mayo deployed the NVIDIA DGX SuperPOD, a supercomputing cluster powered by the new Blackwell architecture.⁴⁷ This infrastructure allows them to train massive foundation models in-house.
• Impact: This compute power enabled the development of a generative pathomics model that reduced the time required to analyze pathology slides from four weeks to one week.⁴⁷ Furthermore, their “Atlas” foundation model, trained on 1.2 million histology images, serves as a base upon which other clinical tools are built.
• Seeing the Unseen: Leveraging this infrastructure, Mayo deployed a pancreatic cancer detection model that can identify malignancy on CT scans taken months before symptoms appear, achieving 97% accuracy by analyzing subtle textural patterns invisible to the human eye.⁴⁹

7.2 Cleveland Clinic: The Hybrid Quantum Approach

Cleveland Clinic has focused on a “Center-Led” governance model and the integration of diverse computing paradigms.

• Governance: They established a center-led model where a central AI office sets the standards (data, infrastructure, ethics), but the specific use cases are driven by clinical institutes.⁵⁰ This ensures that AI solves real clinical problems rather than being a “solution looking for a problem.”
• Quantum AI: Unique to Cleveland Clinic is the integration of the IBM Quantum System One.⁵¹ In 2025, they are exploring hybrid workflows where quantum computers handle complex molecular simulation tasks while generative AI handles data synthesis, creating a pipeline for accelerated biomedical discovery.
• Operational Scale: Their deployment of the “AI Scribe” (Ambience Healthcare) to over 4,000 clinicians is one of the widest in the industry, driven by a philosophy of “restoring the human connection”.³⁴

8. Conclusion: The Agentic Future

As the analysis of 2025 demonstrates, the “Generative Era” of medical AI is no longer a speculative future; it is the operational present. The industry has successfully navigated the “trough of disillusionment” and entered the “slope of enlightenment.”

We have witnessed the birth of Genomic Foundation Models like Evo 2 that turn biology into a computable language, promising to cure diseases at their source. We have seen Agentic AI begin to dismantle the administrative bureaucracy that stifles clinical care, returning time to the physician-patient relationship. And we have seen the establishment of robust Regulatory Frameworks like the FDA’s PCCP and the EU AI Act that provide the guardrails necessary for safe scaling.

However, the “human” element remains the critical variable. As the National Academy of Medicine’s 2025 report concludes, “Knowing is not enough; we must apply”.⁴⁶ The challenge for the remainder of the decade will be organizational, not technological. Success will depend on the ability of health systems to retrain their workforce to partner with agents, to redesign workflows around predictive rather than reactive care and to ensure that the benefits of this “design room” for biology are shared equitably across the global population.

The technology is ready. The governance is emerging. The burden of proof now shifts to the implementation.

Appendix: Key Data Summary (2025)

Table 2: Clinical and Scientific Performance Metrics

Model/Tool	Domain	2025 Performance Metric	Source
Med-Gemini	Clinical Reasoning	91.1% accuracy on MedQA (USMLE-style).	21
Claude 3.7 / GPT-4	Complex Diagnosis	~45-48% accuracy on complex NEJM vignettes.	26
Evo 2	Genomics	Zero-shot prediction of disease-causing mutations; 131kb context.	11
Ambient Scribes	Workforce	Reduced burnout from 51.9% to 38.8% in 30 days.	32
Pancreatic AI	Oncology	97% accuracy in early detection from pre-symptomatic CTs.	49

Table 3: Regulatory Milestones of 2025

Regulation/Guidance	Authority	Key Mechanism	Source
PCCP Guidance	FDA (USA)	Allows pre-approved modification plans for AI models to evolve without new submissions.	5
EU AI Act	EU Commission	Classifies medical AI as “High Risk”; requires fundamental rights impact assessments.	41
MDCG 2025-6	EU MDCG	Clarifies “Provider” (Manufacturer) vs. “Deployer” (Hospital) roles in MDR/AI Act overlap.	42
LMM Guidance	WHO	40+ recommendations for the governance of Large Multi-Modal Models as socio-technical systems.	43

Works cited

1. The state of AI in 2025: Agents, innovation and transformation – McKinsey, https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
2. 2025 AI Trends in Healthcare & Life Sciences | Key Insights – RSI Security, https://blog.rsisecurity.com/trends-in-healthcare-life-sciences/
3. New AI breakthrough can model and design genetic code across all domains of life, https://engineering.berkeley.edu/news/2025/02/new-ai-breakthrough-can-model-and-design-genetic-code-across-all-domains-of-life/
4. 2025: The State of AI in Healthcare | Menlo Ventures, https://menlovc.com/perspective/2025-the-state-of-ai-in-healthcare/
5. Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions | FDA, https://www.fda.gov/regulatory-information/search-fda-guidance-documents/marketing-submission-recommendations-predetermined-change-control-plan-artificial-intelligence
6. The AI Act Explorer | EU Artificial Intelligence Act, https://artificialintelligenceact.eu/ai-act-explorer/
7. Fact check: Is Golden State Warriors star Al Horford permanently out of NBA 2025- 26 due to major health, https://timesofindia.indiatimes.com/sports/nba/top-stories/fact-check-is-golden-state-warriors-star-al-horford-permanently-out-of-nba-2025-26-due-to-major-health-issue/articleshow/125797279.cms
8. FDA Expands Artificial Intelligence Capabilities with Agentic AI Deployment, https://www.fda.gov/news-events/press-announcements/fda-expands-artificial-intelligence-capabilities-agentic-ai-deployment
9. United States Business Intelligence Software Market Analysis 2025-2033 by Technology, Deployment, Enterprise, End User, States and Competitive Landscape, https://www.globenewswire.com/news-release/2025/12/03/3199183/0/en/United-States-Business-Intelligence-Software-Market-Analysis-2025-2033-by-Technology-Deployment-Enterprise-End-User-States-and-Competitive-Landscape.html
10. Generative AI tool marks a milestone in biology | Stanford Report, https://news.stanford.edu/stories/2025/02/generative-ai-tool-marks-a-milestone-in-biology-and-accelerates-the-future-of-life-sciences
11. Sequence modeling and design from molecular to genome scale with Evo – PubMed Central, https://pmc.ncbi.nlm.nih.gov/articles/PMC12057570/
12. Study Finds Generative AI Outperforms Nature in Protein Design | BioPharm International, https://www.biopharminternational.com/view/study-finds-generative-ai-outperforms-nature-in-protein-design
13. AlphaGenome: AI for better understanding the genome – Google DeepMind, https://deepmind.google/blog/alphagenome-ai-for-better-understanding-the-genome/
14. AlphaFold: Five Years of Impact – Google DeepMind, https://deepmind.google/blog/alphafold-five-years-of-impact/
15. AlphaFold 3: an unprecedent opportunity for fundamental research and drug development | Precision Clinical Medicine | Oxford Academic, https://academic.oup.com/pcm/article/8/3/pbaf015/8180385
16. Rational drug design with AlphaFold 3 – Isomorphic Labs, https://www.isomorphiclabs.com/articles/rational-drug-design-with-alphafold-3
17. AlphaFold3 in Drug Discovery: A Comprehensive Assessment of Capabilities, Limitations and Applications | bioRxiv, https://www.biorxiv.org/content/10.1101/2025.04.07.647682v1
18. Novel molecules from generative AI to phase II – EurekAlert!, https://www.eurekalert.org/news-releases/1037177
19. Nature Communications | Insilico Medicine Presents AI-empowered Dual-Action PROTAC Targeting PKMYT1 – PR Newswire, https://www.prnewswire.com/news-releases/nature-communications–insilico-medicine-presents-ai-empowered-dual-action-protac-targeting-pkmyt1-302630345.html
20. Combining AI and new genomic techniques to ‘fine-tune’ plants: challenges in risk assessment – Frontiers, https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2025.1677066/full
21. Advancing healthcare and scientific discovery with AI – Google Blog, https://blog.google/technology/health/google-research-healthcare-ai/
22. Advancing medical AI with Med-Gemini – Google Research, https://research.google/blog/advancing-medical-ai-with-med-gemini/
23. [2404.18416] Capabilities of Gemini Models in Medicine – arXiv, https://arxiv.org/abs/2404.18416
24. (PDF) Advancing Multimodal Medical Capabilities of Gemini – ResearchGate, https://www.researchgate.net/publication/393472202_Advancing_Multimodal_Medical_Capabilities_of_Gemini
25. Comparative analysis of large language models in clinical diagnosis: performance evaluation across common and complex medical cases | JAMIA Open | Oxford Academic, https://academic.oup.com/jamiaopen/article/8/3/ooaf055/8161131
26. Knowledge-Practice Performance Gap in Clinical Large Language Models: Systematic Review of 39 Benchmarks – Journal of Medical Internet Research, https://www.jmir.org/2025/1/e84120
27. Evaluating Large Language Model Diagnostic Performance on JAMA Clinical Challenges via a Multi-Agent Conversational Framework | medRxiv, https://www.medrxiv.org/content/10.1101/2025.08.20.25334087v1
28. Large Language Artificial Intelligence Models and Clinical Reasoning: The Frontier in 2024, https://www.jwatch.org/na58220/2025/01/02/large-language-artificial-intelligence-models-and-clinical
29. Generative Artificial Intelligence in Healthcare: Applications, Implementation Challenges and Future Directions – MDPI, https://www.mdpi.com/2673-7426/5/3/37
30. Generative AI Healthcare: 15 Use Cases with Examples – Research AIMultiple, https://research.aimultiple.com/generative-ai-healthcare/
31. Radiology: Artificial Intelligence – RSNA Journals, https://pubs.rsna.org/journal/ai
32. Use of Ambient AI Scribes to Reduce Administrative Burden and Professional Burnout, https://pmc.ncbi.nlm.nih.gov/articles/PMC12492056/
33. AI Scribes in Health Care: Balancing Transformative Potential With Responsible Integration, https://medinform.jmir.org/2025/1/e80898
34. Less Typing, More Talking: AI Reshapes Clinical Workflow – Consult QD, https://consultqd.clevelandclinic.org/less-typing-more-talking-how-ambient-ai-is-reshaping-clinical-workflow-at-cleveland-clinic
35. Using Generative Artificial Intelligence (AI) to Create Patient-Friendly Discharge Summaries | Clinical Research Trial Listing – CenterWatch, https://www.centerwatch.com/clinical-trials/listings/NCT06711458/using-generative-artificial-intelligence-ai-to-create-patient-friendly-discharge-summaries
36. AI-generated patient-friendly discharge summaries to empower patients – medRxiv, https://www.medrxiv.org/content/10.1101/2025.07.04.25330804v1.full.pdf
37. Time to Study Implementation of AI-Generated Discharge Summaries – PubMed, https://pubmed.ncbi.nlm.nih.gov/40323624/
38. CVS Health (NYSE: CVS) reports $474B U.S. economic impact, 1.3M jobs supported, https://www.stocktitan.net/news/CVS/cvs-health-generates-over-474-billion-in-2024-u-s-economic-impact-5p67zn9b4qfq.html
39. FDA Guidance on AI-Enabled Devices [2025] – Greenlight Guru, https://www.greenlight.guru/blog/fda-guidance-ai-enabled-devices
40. Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations | FDA, https://www.fda.gov/regulatory-information/search-fda-guidance-documents/artificial-intelligence-enabled-device-software-functions-lifecycle-management-and-marketing
41. Navigating the EU AI Act: A Healthcare Compliance Guide | RiskImmune Blog, https://riskimmune.ai/blog/navigating-the-eu-ai-act-a-healthcare-compliance-guide
42. AIB 2025-1 MDCG 2025-6 Interplay between the Medical Devices Regulation (MDR) & In vitro Diagnostic Medical Devices – Public Health – European Commission, https://health.ec.europa.eu/document/download/b78a17d7-e3cd-4943-851d-e02a2f22bbb4_en?filename=mdcg_2025-6_en.pdf
43. Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models, https://www.who.int/publications/i/item/9789240084759
44. WHO releases AI ethics and governance guidance for large multi-modal models, https://www.who.int/news/item/18-01-2024-who-releases-ai-ethics-and-governance-guidance-for-large-multi-modal-models
45. Capturing the Potential of Generative AI’s Use in Health and Medicine Requires Collaboration and Oversight, Consideration of Risks, Says NAM Special Publication, https://www.nationalacademies.org/news/capturing-the-potential-of-generative-ais-use-in-health-and-medicine-requires-collaboration-and-oversight-consideration-of-risks-says-nam-special-publication
46. Generative Artificial Intelligence in Health and Medicine: Opportunities and Responsibilities for Transformative Innovation (2025), https://www.nationalacademies.org/read/28907/chapter/1
47. Mayo Clinic: New AI Computing Platform Will Advance Precision Medicine | AHA, https://www.aha.org/aha-center-health-innovation-market-scan/2025-08-12-mayo-clinic-new-ai-computing-platform-will-advance-precision-medicine
48. Mayo Clinic deploys NVIDIA Blackwell infrastructure to drive generative AI solutions in medicine, https://newsnetwork.mayoclinic.org/discussion/mayo-clinic-deploys-nvidia-blackwell-infrastructure-to-drive-generative-ai-solutions-in-medicine/
49. Seeing the Unseen – Mayo Clinic Magazine, https://mayomagazine.mayoclinic.org/2025/06/seeing-the-unseen/
50. Building and Implementing an Artificial Intelligence Action Plan for Health Care | American Hospital Association, https://www.aha.org/system/files/media/file/2025/01/Market_Insights_AI_Report-2025.pdf
51. Cleveland Clinic and G42 to Advance Healthcare through Artificial Intelligence, https://newsroom.clevelandclinic.org/2025/03/25/cleveland-clinic-and-g42-to-advance-healthcare-through-artificial-intelligence