Community oncology programs do not have to wait for a formal AI strategy to encounter AI in their daily work. Staff may already be using generative AI to draft emails, summarize information, organize schedules, develop first drafts of policies, or cut down on administrative work. At the same time, many practices are still deciding what staff are allowed to use, what information can be entered, who should approve new tools, and how closely AI outputs need to be checked.

New resources from the Association of Cancer Care Centers (ACCC) offers community oncology programs a practical place to start. The materials—consisting of two information guides, a checklist, and two on-demand webinars—cover day-to-day use of generative AI, readiness questions to ask before adopting an AI-enabled tool, survey findings on how cancer programs are already using AI, and governance gaps that can appear when adoption moves faster than policy.

Adam P. Dicker, MD, PhD, FASTRO, FASCO, an ACCC committee member and Chair and Professor of Radiation Oncology, Pharmacology, and Experimental Therapeutics at Thomas Jefferson University, said ACCC was trying to answer a question many practices are facing now: not just why AI matters, but how to begin using it. “There are lots of articles explaining the why,” Dr. Dicker said. “We focused on how to enter the workflow and how to make this manageable.”

Practices also need the infrastructure, training, and workflows to use the technology well, he said, which can be harder for smaller community practices than for larger systems. Community practices need “the basics and an introduction to where to start,” Dr. Dicker said, adding that the goal was to “make it manageable, realistic, and attainable to all levels of care.”

Getting Started With Generative AI in Cancer Care Delivery and Operations

This practical tip sheet is for oncology teams beginning to use tools such as Microsoft Copilot, ChatGPT, Google Gemini, and Claude in routine work. It is less a technical manual than a safe-use primer: what generative AI can reasonably help with, what it should not be used for, and what guardrails need to be in place.

The resource focuses on lower-risk, everyday uses over high-stakes clinical decision-making. These include drafting outlines or policies, creating phone scripts, summarizing registry data, and helping with operational tasks such as infusion staffing or mobile lung screening schedules.

The advice is direct: protect sensitive information, use AI for a first draft rather than a final answer, start with low-risk tasks, and check the output carefully. The resource also tells users to be specific in their prompts, give context, define the desired tone or format, and ask follow-up questions.

Andrew Norden, MD, MPH, MBA, an ACCC committee member, board-certified neuro-oncologist, and physician executive serving as Chief Medical Officer at OncoHealth, a digital health and oncology analytics company, said practices should not be intimidated by AI, but they should understand its limits.

“Generative AI tools can transform the work of nearly every professional involved in cancer care and management,” he said. “They offer remarkable benefits across an array of tasks, and people should not be afraid to use them in the clinic. That said, it is critical to ensure compliance with organizational policies and data privacy rules, and to keep in mind that output accuracy is not 100%; human review remains a requirement for all but the lowest risk use cases.”

The resource warns users not to enter protected health information or confidential business information into tools that have not been approved for that purpose. It also tells users to verify outputs, check sources, assume mistakes may be present, and avoid using AI for tasks that require clinical judgment or decision-making unless the tool is approved and regulated for that use.

ACCC frames generative AI as a starting point rather than a final step. The resource encourages teams to use the tools for drafts, summaries, outlines, and other lower-risk work, then review the outputs against internal policies, source materials, and clinical or operational judgment.

Ian Miller, an ACCC committee member and Associate Director at the Digital Medicine Society (DiMe), said oversight should stay front and center as practices use generative AI. “Maintaining clinical and human oversight of the tools is paramount for patient safety and overall durable, sustainable implementation,” Miller said.

For leaders, the resource says broad AI policies may need to be turned into clear oncology-specific rules: which tools staff can use, what information should never be entered, which uses are allowed, who checks the output, and where staff should go with questions. It also recommends providing training for the staff about how to prompt and evaluate outputs to ensure safe AI use. 

Assessing AI Readiness in Oncology

This companion checklist is for leaders deciding whether to adopt an AI or automation tool. It starts with a basic question: is a new tool needed at all?

The checklist asks practices to identify the operational or care delivery problem they are trying to improve, who is affected, where the problem occurs in the cancer care continuum, and whether the need is coming from frontline staff, outside pressure, vendor interest, or general enthusiasm about new technology.

One of its core tips is: “Start with the problem, not the technology.” From there, the checklist asks practices to look at fit. Do current systems already have features that could address the problem? Could workflow changes solve part of it? Would an AI-enabled solution duplicate something already working well, or add another screen, login, or process to staff who are already stretched? Dr. Norden said this is one of the steps practices can miss.

“One key step that programs often skip is assessment of fit with current tools and workflows,” he said. “Duplicating an existing, effective platform with an AI-enabled solution may offer limited benefits, while solving a thorny problem that is well-suited to a novel AI system could be particularly valuable.”

The checklist also asks leaders to define success before implementation. ACCC suggests identifying baseline data and tracking measures before a pilot begins, including operational metrics such as time to first treatment, infusion chair use, and symptom triage response time, along with documentation time, patient experience, accuracy, and staff workload.

Miller said several foundations should come before adoption, including monitoring and performance safeguards, education for clinicians or end users, and adequate data and technology infrastructure. “These foundations are really what make or break a successful implementation,” he said.

The checklist also makes clear that AI adoption takes more than a software purchase. Practices are asked to think about training, integration, workflow changes, staffing adjustments, safety monitoring, pathway alignment, protocol updates, and quality or accreditation needs. If a tool changes the way staff document, triage, schedule, navigate, or communicate with patients, the practice has to plan for that work.

The resource also asks practices to assign ownership. ACCC recommends a one-page AI use guidelines document, a multidisciplinary oncology review group, a primary point of contact for questions or escalation, and early engagement with the staff who will actually use the tool.

The checklist ends with equity, safety, and regulatory questions. Practices are encouraged to ask whether a tool could introduce bias or worsen disparities, whether equity-related outcomes should be tracked, and whether regulatory, accreditation, or reporting requirements need to be considered before use expands.

Dr. Dicker said this kind of implementation guidance is especially important for community practices. A health system may understand why technology could help, but that does not mean it has the workflows, training, or infrastructure to use it well.

AI in Cancer Care Delivery and Operations

An ACCC member survey collected responses from 168 volunteers at 102 unique cancer centers between May and August 2025. Fifty-five percent of respondents said they were experimenting with AI tools “unofficially” to support parts of their role, and most reported at least some hands-on professional experience with AI. The findings show that AI is not just a future issue for cancer programs.

At the same time, confidence was limited in areas that matter for safe adoption. ACCC reported that average confidence scores were neutral or below neutral across questions about recognizing AI limitations, explaining how AI works and how it can be used in cancer care, and evaluating AI systems. The concern is that staff may already be using AI before training and policy have caught up.

The survey also found uncertainty inside organizations. Depending on the category, many respondents were unsure whether their programs were implementing, planning, considering, or not using AI tools in clinical, administrative, operations, patient engagement, or research functions. If a program does not know which tools are being used, it cannot easily monitor them, train staff on them, or decide whether they should be expanded.

Dr. Norden said the biggest concerns ACCC is hearing from community oncology programs relate to “the unknown impact on clinical judgment as clinicians increasingly rely on AI to support clinical decisions” and “the risk that AI could be used to make critical care decisions without human oversight.” By contrast, he said, there are “few concerns about the use of AI to support non-clinical and operational use cases.” For practices just beginning to evaluate AI, the materials point first to lower-risk operational uses. Clinical uses require more caution, oversight, and accountability.

The survey also asked about barriers and needed supports. Respondents named cost, lack of guidelines and regulation, and data privacy and security as top implementation barriers. Evidence and effectiveness, engagement and buy-in, and hands-on training and support were the top factors they said would support implementation.

The AI That’s Already Here: Promise, Risk, and the Governance Gap

This report builds on the survey findings and focuses on what happens when AI use moves faster than formal oversight. The report’s starting point is blunt: AI is already being used in oncology. The question is whether programs have the guardrails to manage it.

ACCC reported that 55% of survey respondents were using AI tools their organizations had not officially approved, compared with 45% who were using organizationally sanctioned tools. The report argues that this kind of unofficial use can expose gaps in governance.

Those gaps may include a lack of multidisciplinary review, formal evaluation before clinical deployment, centralized inventories of AI tools, equity-impact review, or clear accountability for AI-related errors. Without that structure, programs may not know which tools staff are using, what information is being entered, who is checking the output, or who is responsible if something goes wrong.

The ACCC report recommends starting with the workflow rather than the tool and testing tools in small pilots before scaling. It also emphasizes champions, training, safe places for staff to practice, outcome measurement, and monitoring for unintended consequences.

The report also emphasizes clear, written communication so staff are not left to make their own decisions about AI use. One focus group recommendation was simple: “Write down what is allowed.” That means spelling out which tools staff may use, what tasks they may use them for, and what information should never be entered into an AI system.

Dr. Norden said that written guidance should be paired with multidisciplinary oversight. “One of the most important steps a program should take is to create a multidisciplinary oversight group comprised of subject matter experts in clinical oncology, operations, IT, and regulatory compliance,” he said.

That group, he added, should have real authority. “This entity should be empowered to evaluate new AI use cases, monitor implementation, and halt efforts in case of safety or other serious concerns,” Dr. Norden said.

Miller also emphasized the need for ongoing monitoring. He pointed to ambient scribes as one example of AI tools that may reduce administrative burden, but said such tools can also hallucinate. “Patient safety and privacy should always remain top priority, so we’ve learned that these types of tools necessitate ongoing monitoring and human oversight to protect from unintended consequences,” Miller said.

He added that practices may get more value from AI when they understand the technology more deeply rather than treating it as a plug-in fix. “We also see that the more intimately a practice or system interacts and understands the technology, the more value they can extract,” Miller noted. “So, ultimately, responsible implementation is what makes efficiency gains possible and sustainable.”

Dr. Dicker raised a related point about technologies that are “listening” in clinical settings. He said that can be both a benefit and a hindrance in oncology practice, and that ACCC hopes to focus more on best practices in future phases of its AI programming.

The ACCC was not trying to add another broad argument for AI, he added. Instead, the resources are meant to help practices move from interest in AI to the practical questions that determine whether a tool can actually be used safely and effectively.

ASCO AI in Oncology is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (ASCO^®). The ideas and opinions expressed in ASCO AI in Oncology do not necessarily reflect those of Conexiant or ASCO. For more information, see Policies.