On November 5, 2026, the University of Maryland School of Medicine will host a one-hour Zoom seminar titled “AI Tools in Research: Navigating Ethical Risk and Research Integrity,” led by Henry Silverman, MD, MA, HEC-C. According to the UMSOM events calendar, the session runs from 9:00 to 10:00 a.m., is organized by the Center for Advanced Research Training and Innovation (CARTI), and requires registration. It lands at a tense moment for AI in laboratories and clinics, when AI research ethics is no longer a policy slogan but a daily decision for anyone touching data, code, or manuscripts.
UMSOM’s hour on AI research ethics: who, when, where
Silverman, a professor of medicine and certified bioethicist, will guide the discussion as researchers reckon with new tools that draft protocols, summarize literature, and even propose hypotheses. The calendar listing states the event is virtual via Zoom and open by registration, signaling demand across departments that may be piloting AI plugins and pipelines without shared guardrails. For labs that have adopted chat-based assistants or automated summarizers to speed reviews and analysis, this is a chance to turn scattered norms into clear practice, with research integrity front and center.
Why real incidents put research integrity on the line
Concerns are not abstract. On August 31, 2026, The Guardian reported that an NHS watchdog warned doctors’ AI scribes were getting drug names and diagnoses wrong, raising safety and reliability questions in clinical documentation (The Guardian AI coverage). Two days earlier, the outlet highlighted research showing a sharp rise in incidents where AI systems escaped users’ intended control. Neither story was about academic labs per se, yet both map directly to risks in scholarly work: unverified model outputs that creep into records or drafts, and tools that behave unpredictably when pushed beyond tested scenarios.
These failure modes intersect with long-standing expectations in science. Data provenance must be traceable. Methods must be reproducible. Authorship must reflect real intellectual contribution. When a model rewrites an introduction or drafts a data table, who is accountable for errors or invented citations? When a pretrained model absorbs sensitive data during fine-tuning, who can attest to privacy protections and downstream access? The incidents The Guardian surfaced show how small lapses cascade into large integrity problems.
What questions labs should bring to the Zoom
Turning AI research ethics into action starts with concrete prompts that PIs, postdocs, and students can use the next time a model window opens. This seminar is an opportunity to press for specifics. Useful questions include:
- What documentation should accompany any AI-assisted analysis or writing, so that peers can audit where and how a tool influenced results?
- Which tasks in our lab are acceptable for AI assistance, which are prohibited, and who approves exceptions?
- How do we validate AI-generated outputs against ground truth before they enter manuscripts, protocols, or clinical records?
- What is our process for handling sensitive data with third-party AI services, including consent, de-identification, and vendor terms?
- How do we assign authorship or acknowledgments when AI tools contribute text or figures, and how do we disclose this to journals?
Expectations exist beyond the lab. The U.S. Office of Research Integrity offers practical guidance on responsible conduct that can anchor these choices; its resources on data management, authorship, and peer review remain a useful baseline (ORI). The National Academies’ report “Fostering Integrity in Research” sets out norms for transparency and accountability that apply cleanly to AI-assisted workflows (National Academies). Those principles translate well to model prompts, logs, and citations.
From policy to practice: ethical AI in research, every day
Policies can read clear on paper, then collapse in the rush of a deadline. The habits that prevent that collapse are simple and tedious, which is why they work. Keep a model log: date, task, prompts, model version, and any manual edits. Freeze versions for analyses, the same way you would pin a library or container. Ban paste-and-go for references; verify citations manually or with a database query before a draft leaves the lab. Never allow sensitive or unpublished data into consumer chatbots without a written, reviewed plan that covers consent and vendor controls.
Most of all, test. Before an AI tool touches a high-stakes task, run it on a validation set you understand well. If it helps, retain it; if it harms, retire it. The Guardian’s reporting on errant AI scribes shows why this matters in clinics. The same discipline applies to bench and computational science. Small trials at low risk, measured against clear baselines, limit the blast radius when a tool behaves in unexpected ways.
Journals and funders are aligning to these expectations. Many venues encourage or require disclosures when AI tools assist with writing or figure generation, and some specify that AI systems cannot be listed as authors because they cannot accept responsibility. While disclosure formats vary, the direction is stable: document the role of tools, name the humans who are accountable, and preserve the artifacts needed for review. That is research integrity, translated for the age of models.
How to get the most from the UMSOM session
Come with a short list of live use cases in your group: literature screening, code debugging, statistical summaries, or patient-note structuring. Ask where each sits in your institution’s policy, what documentation is required, and how risk shifts if work moves from a local model to a cloud API. AI research ethics becomes practical when tied to real workflows and deadlines.
The seminar’s focus on ethics and integrity is timely, and it is designed to be usable. According to the UMSOM calendar, CARTI is hosting to support researchers across the School of Medicine. With incidents piling up in the headlines and lab tools changing by the month, one disciplined hour can prevent months of rework or, worse, a retraction.
Registration is required. If your work touches data, methods, or manuscripts, you’ll want a seat. AI research ethics isn’t a box to check; it’s how the work stands up when someone else reads it, reruns it, and trusts it. For more on this, see bloomberg.com and nytimes.com.