Tag: artificial-intelligence

Modified on March 5, 2026 to remove the “subscribe” option. This blog has been retired and replaced by the S.P.I.R.I.T. newsletter.

Good morning, good afternoon, and good evening, Compliance Rockstars, Clinical Research Professionals, Ethics Enthusiasts, Legal Experts, and Investigators!

300+ subscribers and counting!

It’s mind blowing what’s transpired in 2025 so far (and we’re barely over the midpoint). I don’t solely mean on a personal level. So much has happened from a regulatory standpoint. It’s certainly difficult to keep up with everything. Life feels like it’s moving so fast.

Through my blog, I hope you feel this is a place where you can slow down and catch up. This blog is meant to serve you as an educational resource on-the-go (and at rest). I always welcome (and promote) feedback via comment on the blog or email: crest.innovation25@gmail.com

Today’s post will cover breakthroughs made in artificial intelligence (AI) in human research, healthcare, and medicine both domestically and internationally. AI research on the rise with no end in sight. It is essential to familiarize yourself with current best practices from industry experts.

I’d like to take a moment to thank these dedicated individuals who have devoted themselves to these working groups.

I am fortunate to know some of these folks through personal connection and applaud you. For those I don’t, I am grateful that you have willingly shared your expertise and time for this tremendous effort.

All your voices collectively are needed now more than ever. We are in an arms race with AI innovation and promoting the ethical conduct and use of AI in research.

As a general reminder, these are my own interpretations. Any legal information discussed within this post should be discussed with your institution or organization.

Let’s commend and review current progress in AI frameworks:

World Health Organization’s (WHO) Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models

You can review the guidance here: Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models

The WHO guidance begins with an executive summary highlighting the following points:

• The brief introduction of the concept of AI and large multi-model models (LMMs) are introduced along with their initial work with this in 2021: Ethics & Governance of Artificial Intelligence for Health
  • From this initial guidance, potential risks and benefits were identified for the use of AI in healthcare.
  • Further, the following ethical principles were noted for governments, public sector agencies, researchers, companies, and implementers to consider:
    • Protect autonomy
    • Promote human well-being, human safety and the public interest
    • Ensure transparency, explainability and intelligibility
    • Foster responsibility and accountability
    • Ensure inclusiveness and equity
    • Promote AI that is responsive and sustainable
• Through the current WHO guidance, it will:
  • Assist in mapping the benefits and challenges associated with use of LMMs for health and in developing policies and practices for appropriate development, provision and use
  • Provide recommendations for governance, within companies, by governments and through international collaboration, aligned with the guiding principles

The guidance is then broken down into two sections:

• The first section details the applications, challenges, and risks of LMMs
  • Applications of AI for health include:
    • Diagnosis
      • This area is particularly promising given LMMs can see complex or rare cases of a diagnosis
    • Clinical care (along with public health surveillance)
      • With respect to patient-centered applications, AI is revolutionizing how patients are tending to their health (e.g., self-care chatbots and prediction tools).
        • Though is NOT mentioned in this section, I wanted to share the downside of health chatbots. Particularly in therapy (article published in June 2025): Therapy Chatbot Tells Recovering Addict to Have a Little Meth as a Treat
    • Research and drug development
      • I’m sure we’ve seen plenty of examples of this. However, the example mentioned here is the review of electronic health records to identify current clinical practice patterns. This could lead to the development of a new clinical practice
    • Healthcare administration
      • Here, AI could be used to reduce administrative burden (which could lead to a reduction in employee burnout)
  • Risk of using these LMMs include:
    • Overestimating benefits and underestimating risks
    • Accessibility and affordability
    • System-wide biases (such as exclusion of particular populations)
    • Impact on employment (likely due to AI automation)
    • Dependence of health systems on unsuitable LMMs (i.e., LMMs that are not maintained regularly)
    • Cybersecurity risks (e.g., malicious hacking)
  • Challenges noted for the use of these LMMs include:
    • Financial burden (e.g., having computers run continuously as well as training and deployment)
    • Attrition within universities and government agencies
      • The guidance highlights how faculty members who specialize in AI are being hired away from universities to work in industry (as industry LMMs are much larger and obtain greater investment towards the tool)
    • Lack of corporate commitment to ethics (given the elimination of employees dedicated to this effort)
      • The pressure to generate and maintain these LMMs tend to push ethics to the waste side
    • Lack of a moral compass (again, please read the article mentioned earlier about therapy chatbots)
• The second section details the ethics and governance of LMMs in healthcare and medicine
  • Recommendations are provided to LMM developers for implementation to address these ethical risks:
    • Hiring individuals with AI expertise in science and engineering (e.g., via certification)
    • Ensure high data quality in training these LMMs as well as strict adherence to laws of informed consent and other privacy regulations (such as GDPR)
    • Design with human rights and values in mind (e.g., inclusiveness and transparency)
      • Environmental concerns should also be considered (e.g., carbon footprint and water usage)
  • With respect to governance:
    • Governments “should have clear data protection laws and regulations for the use of health data and protecting individual rights, including the right to meaningful informed consent”
    • Governments should have design and development standards as well as audits throughout LMM development
    • 20 recommendations were listed with respect to open-source LMMs

UNESCO’s Red Teaming artificial intelligence for social good – The PLAYBOOK

You can review the guidance here: Red Teaming artificial intelligence for social good – The PLAYBOOK

UNESCO’s playbook begins with a short summary highlighting its intended use. The primary concept described is using Red Teaming for evaluating Gen AI systems for social good and exposing harms.

• Red Teaming is described as a hands-on exercise where participants test Gen AI models for flaws and vulnerabilities that could unveil harmful behavior
  • A graphic is used to describe Red Teaming in four steps:
    • Find weaknesses in AI systems that could lead to errors, vulnerabilities, or bias
    • Set safety benchmarks
    • Collect diverse stakeholder feedback
    • Ensure models perform as expected
• The playbook can be used for a vast range of professionals ranging from researchers to nonprofits to artists
• When performing and preparing for Red Teaming exercises:
  • It’s essential to know the difference between unintended consequences and intended malicious attacks
  • Teams should have:
    • Clear objectives (i.e., defined challenge and prompts),
    • A diverse group of team members, and
    • Choose the appropriate format to conduct these exercises
• Once exercises are complete, the playbook has the following recommendations when interpreting results:
  • Stay focused on team’s hypothesis
  • Avoid jumping to conclusions
  • Use different analytical tools for different sized datasets
• The playbook also discusses potential challenges for implementing these exercises and how to overcome them

National Academy of Medicine’s (NAM) An Artificial Intelligence Code of Conduct for Health and Medicine: Essential Guidance for Aligned Action

You can review the guidance here: An Artificial Intelligence Code of Conduct for Health and Medicine: Essential Guidance for Aligned Action

Similar to the WHO guidance, the NAM AI Code of Conduct (AICC) opens with an executive summary:

• The objective of the AICC is to harmonize the existing principles, address identified gaps, and map these principles to the NAM’s Learning Health System (LHS) Shared Commitments
• The AICC framework highlights six commitments:
  • Advance humanity
  • Ensure equity
  • Engage impacted individuals
  • Improve workforce well-being
  • Monitor performance
  • Innovate and learn
• The summary also highlights key stakeholder groups who contributed to the development of the AICC
  • Perspectives of these key stakeholders are described in great detail later in the AICC with respect to the six commitments
    • Common themes between these groups and distinct contributions from each group are also included

The framework continues with providing additional background information such as:

• Defining AI and how it differs from other rule-based digital health technologies
• Describing the use of AI in health, healthcare, and biomedical sciences
• Risks associated with AI use and challenges AI use poses on governance and regulations

The AICC principles were updated based on public comment and NAM working group feedback. These principles are described in Table 3-2 in detail and are briefly listed here:

• Engaged
• Safe
• Effective
• Equitable
• Efficient
• Accessible
• Transparent
• Accountable
• Secure
• Adaptive

It is noted how these AICC principles and commitments can be applied to the AI life cycle. The AICC concludes with how these principles and commitments can potentially be regulated in a tight-loose-tight framework.

The Multi-Regional Clinical Trials (MRCT) Center of Brigham and Women’s Hospital and Harvard and WCG’s Framework for Review of Clinical Research Involving AI

You can review the guidance here: Framework for Review of Clinical Research Involving AI

This collaborative framework provides guidance to IRBs (and other reviewing committees) with actionable steps in reviewing AI research. Specifically, how to identify, assess, and mitigate risks to participants. The framework (also known as the toolkit) is broken down into logical sections:

• A decision tree in determining if IRB review is required for the proposed AI research project with respect to the Common Rule and FDA regulations
• A guide of questions and considerations to assess AI tool development and training data used in three phases:
  • Discovery
  • Translation
  • Deployment including:
    • Algorithm stability
    • Data identifiability, sources, and collection
• Questions to consider with respect to the following ethical principles:
  • Human agency and oversight
  • Technical robustness and safety
  • Privacy, confidentiality, and data governance
  • Transparency
  • Representative and fairness
  • Informed consent
• Finally, the toolkit provides a checklist when considering the use of AI in the administration of research. Though this may fall out of the IRB’s purview, examples of these types of activities include:
  • AI-enhanced data analysis
  • Human subject recruitment
  • Use of LLMs to help develop:
    • Protocols
    • Subject facing materials (e.g., informed consent forms, or recruitment materials)
    • Research instruments (e.g., questionnaires, data collection tools)
  • Transcription of interviews and generation of transcripts
  • LLM-generated responses to participant questions about the research
  • Any other operational roles where AI is not the primary intervention

European Medicines Agency’s (EMA) Review of AI/ML applications in medicines lifecycle

You can review the guidance here: Review of AI/ML applications in medicines lifecycle

This brief report highlights AI/ML application publications relevant to future EMA activities. Publications were chosen based on specific inclusion and exclusion criteria (as this wasn’t meant to be a comprehensive review). Challenges and opportunities were explored with respect to:

• Drug discovery
• Nonclinical development
• Clinical trials
• Precision medicine
• Product information
• Manufacturing
• Post-authorization phase

Finally, this report stressed the importance of data protection, data privacy, compliance with regulatory standards, and adapting frameworks to accommodate for the evolution of AI tool use.

I hope you found this summary useful!