Emerging Litigation Podcast

Episode: Agentic AI on Trial: You Be The Judge Part 1 – Medical Diagnostics
Date: January 21, 2026
Host: Tom Hagy
Panelists: Galina Datzkovsky (AI/Compliance Strategist), Marina Kaganovich (Attorney & Google Compliance Advisor), Judge Lisa Walsh (11th Judicial Circuit of Florida)

Episode Overview

This episode inaugurates a special "You Be The Judge" series on the legal and ethical risks posed by agentic AI—AI systems capable of acting autonomously—in high-stakes environments. Focusing on medical diagnostics, the discussion explores emerging litigation issues, liability, standards of care, and workflow design in cases where AI makes diagnostic decisions with minimal or no human oversight. The panel evaluates a hypothetical legal scenario of an autonomous AI agent in mammography, probing who or what might be liable when things go wrong.

Key Discussion Points & Insights

1. Defining Agentic AI and Its Risks

(04:55 – 08:24)

• Galina Datzkovsky provides foundational definitions:
  • Agentic AI refers to systems that can act independently, set and pursue their own goals with little human oversight.
  • Distinguishes agentic AI from "traditional" narrowly focused or rule-based AI.
  • Generative AI enables creativity, but agentic AI takes a further step by enabling proactive, strategic, and autonomous action.

Notable Quote — Galina Datzkovsky (07:03):
“Agentic AI refers specifically to artificial intelligence systems that possess agency, which is really the ability to act independently, make decisions and pursue goals with minimal human oversight ... Agentic AI is proactive, strategic and action oriented.”

2. Hypothetical Scenario: AI in Mammography Screening

(08:24 – 12:16)

• Marina Kaganovich presents a scenario:
  • An AI agent screens mammography scans, forwarding only positive results to a radiologist and directly notifying patients (all-clear or follow-up).
  • If the AI misclassifies a positive as negative, the patient gets a false sense of security and may discover their cancer much later, possibly with a worse prognosis.
  • Raises immediate questions about liability—developers, calibrators, physicians, or institutions?

3. Exploring Potential Liability

(12:16 – 18:14)

• Judge Lisa Walsh unpacks legal concepts of “agency”:
  • In law, "agent" vs. "principal" is narrowly defined—who is holding out authority, who is responsible for actions?
  • Multiple parties could be targets in litigation:
    • AI developer (design/algorithm flaws)
    • Person who calibrates the system (settings, risk sensitivity)
    • Doctor (clinical implementation)
    • Hospital/Health Network (deployment, oversight)
    • Data trainers/providers (quality of training data)
  • Liability depends on who made pivotal decisions and contracts/licensing terms.
  • Agency in legal terms is not a neat match with technical agency.

Notable Quote — Judge Lisa Walsh (13:10):
“Typically what the focus is on is on something called the principal. So the principal ... is whoever it is who holds an agent out as having the authority to do something, and then the agent who does something may bind the principal.”

4. Training Data Quality and System Calibration

(18:14 – 20:21)

• Galina Datzkovsky queries the impact of insufficient or non-diverse datasets—how flaws or biases baked into training affect liability.
• Judge Walsh agrees dataset issues could underpin liability, especially if claims of AI superiority over human review create unjustified trust.

Notable Quote — Judge Lisa Walsh (18:50):
“If the data that the system was trained on is out of date, isn’t sensitive enough, doesn’t alert the system to be able to recognize a malignancy or an abnormality ... that in and of itself might be a ... subset of potential liability because it’s a flawed system.”

5. Standard of Care in the Age of AI

(20:21 – 23:40)

• Panel explores: Is there a unique "standard of care" for machine learning systems versus traditional medical practitioners?
• Medical malpractice law uses professional standards—AI is not a physician.
• Questions arise whether a new, possibly more stringent legal standard will emerge for AI, given expectations of AI’s consistency and lack of human frailty.

Notable Quote — Judge Lisa Walsh (21:38):
“There is a question in my mind as to whether there will be a development in the law of an entirely new body of law of how you evaluate standard of care ... That’s not. It is medical care, but it’s not the standard of professional care of a physician, you know, a human physician or medical technician.”

6. Balancing Efficiency and Oversight

(23:48 – 25:07)

• Marina Kaganovich notes that, to date, even when AI is used, radiologists still review mammograms. Fully automated systems remove both the safety net and efficiency balance.
• Questions: What mitigations (thresholds for review, feedback loops, random checks) can ensure negative results are truly negative?

7. Workflow Mitigations and Quality Control

(25:07 – 29:33)

• Judge Walsh suggests possible safeguards:
  • Multi-point data review before declaring a scan negative.
  • Patient opt-in for human review.
  • Quality control: sample secondary reviews to check system accuracy over time.

Notable Quote — Judge Lisa Walsh (29:33):
“So what about quality control? Even after the system is rolled out, you need to know what is the real result for the way it's working... there has to be some manner of determining that to ensure that it’s doing what it says it should be doing.”

8. Responsibilities of Developers vs. Deployers

(29:58 – 31:54)

• Galina Datzkovsky: Developers should use robust training data, permit tailored calibration, and protect against known AI failure modes (e.g., hallucinations).
• Responsibility shifts: Once a product is purchased, hospitals/providers must conduct onboarding checks, set appropriate thresholds, and ask the right questions during procurement.

9. AI Hallucinations and Prompt Engineering

(31:54 – 32:23)

• Judge Walsh asks whether prompt design or system configuration can help minimize “hallucinations” (AI making things up).
• Datzkovsky clarifies hallucinations are less about prompts in medical imaging, more about model architecture and implementation.

Notable Quote — Galina Datzkovsky (32:23):
“When I mean eliminate hallucinations, you would use the right models ... you’re using those models that will be very precise on the data and wouldn’t be specifically looking to get creative around the data.”

Memorable Moments & Quotes

• Hagy, Opening (01:32): “It’s no longer optional for insurance companies. It’s an imperative to remain competitive. AI now drives claims processing, fraud detection, and even coverage decisions. But speed comes with controversy.”
• Judge Walsh (18:50) on “urban myths”—the public’s trust that AI is better than human review—fuels risk if performance doesn’t live up to expectation.
• Datzkovsky (27:11): “If it goes to a human being, to a doctor, for example, they could miss exactly the same things that the system does. So when you say what does it mean ‘negative is negative,’ you would come back to that idea of training the agent to understand what human beings consider negative.”

Important Timestamps

• 02:11 – 03:02: Guest introductions and episode structure
• 04:55 – 08:24: Galina defines agentic AI versus traditional AI
• 08:24 – 12:16: Hypothetical mammography use case and liability setup
• 12:16 – 18:14: Legal agency and routes to liability
• 18:14 – 20:21: Data quality’s impact on legal exposure
• 20:21 – 23:40: Medical standard of care and AI
• 25:07 – 29:33: Workflow mitigation, thresholds, and quality control
• 29:58 – 32:23: Developers’ responsibility, hallucinations, prompt engineering

Tone

The discussion is collegial yet rigorous. Judge Walsh approaches legal questions analytically but with a clear caveat that the law is evolving in this space. Galina and Marina balance technical and compliance perspectives, frequently looping back to the importance of context, reasonable expectations, and practical best practices for both developers and users.

Summary Takeaways

• Agentic AI is blurring the boundary between tool and independent actor, raising complex liability and ethical questions in medical diagnostics.
• Multiple parties may share liability—developers, calibrators, healthcare providers, and those overseeing implementation—depending on workflows, system design, training data, and oversight practices.
• Expectations of AI performance may lead to higher or new legal standards, especially as systems become more autonomous and claims about their accuracy are made.
• Robust training data, careful calibration, flexible workflows, and ongoing quality control are all critical for legal defensibility and patient safety.
• The law is not yet settled; cases like the one described will shape the future boundaries of responsibility as agentic AI becomes embedded in life-and-death decisions.

End of Summary