Artificial intelligence is no longer knocking on the door of oncology.
It is already inside.
At ASCO 2026, AI was not discussed as a distant promise or a shiny add-on to future cancer care. It was everywhere: clinical decision support, pathology, endoscopy, trial matching, patient information, biomarker discovery, workflow management and precision treatment selection.
That is the shift.
AI is no longer just helping oncology imagine what might be possible. It is starting to shape how cancer care is organised, interpreted and delivered.
For years, the promise has been easy to sell: faster diagnosis, smarter prediction, better trial recruitment, less administrative pressure, more personalised care and wider access to expertise.
But once AI starts influencing clinical judgement, the conversation changes.
Who validates the tool?
Who checks whether it still works six months later?
Who explains the output to the patient?
Who is responsible when AI-supported advice is wrong?
And who makes sure AI does not simply automate the same inequalities that already exist in cancer care?
IPM Take
The AI debate in oncology is no longer only about innovation. It is about power.
That is what makes it political.
AI tools are not neutral once they begin to influence who is flagged as high risk, who gets tested, who is matched to a trial, who receives a treatment recommendation, or who is reassured that nothing urgent is needed.
At that point, AI becomes part of the decision infrastructure of healthcare.
That does not mean oncology should fear AI. It means oncology should govern it.
The risk is not that AI arrives. It already has.
The risk is that it arrives faster than accountability.
From helpful tool to hidden gatekeeper
One of the clearest signals from ASCO was the movement of AI into the space between evidence and decision.
Guideline assistants, clinical trial matching platforms, pathology algorithms and predictive models all promise to help clinicians manage the impossible volume of modern oncology knowledge. That need is real. No oncologist can personally absorb every paper, guideline update, biomarker pathway and trial option across a field moving this fast.
Used well, AI could help reduce variation in care, support overloaded teams and bring high-quality knowledge closer to patients, wherever they are treated.
But cancer policy has to draw a line.
When AI helps clinicians find evidence, it is an information tool.
When AI shapes who is seen as eligible, risky, treatable or trial-ready, it becomes a gatekeeper.
That is where governance matters.
A black-box system that quietly shapes cancer decisions without clear validation, oversight or explanation is not just innovation. It is a new form of authority inside the health system.
And authority needs rules.
The access promise is real. So is the access risk.
AI is often presented as a way to democratise expertise.
In the best-case scenario, it could help clinicians in smaller hospitals access specialist-level guidance. It could support trial matching outside major academic centres. It could improve diagnostic consistency in places where pathology or radiology capacity is stretched. It could help patients understand complex information and prepare better questions for their care team.
That is the optimistic version.
But there is another version too.
AI tools may be trained on data from wealthier systems, validated in narrow populations, bought first by better-resourced hospitals and deployed where digital infrastructure is already strongest.
If that happens, AI will not close the cancer divide.
It will give better tools to systems that are already ahead.
This matters deeply for personalised medicine. Access to molecular testing, digital pathology, high-quality data, genomics and specialist interpretation is already uneven. If AI is built on those same uneven foundations, it may reinforce the gaps it claims to solve.
AI cannot democratise cancer care if the evidence behind it excludes the patients most likely to be left behind.
Validation is now a policy problem
Cancer systems are used to asking whether a medicine works.
They are less prepared to ask whether an algorithm works.
But they need the same discipline.
An AI model that performs well in one hospital may not perform the same way in another. Imaging equipment, pathology workflows, coding practices, population diversity, data completeness, clinical habits and local infrastructure all affect performance.
So AI cannot be treated as a plug-and-play solution.
It needs local validation. External validation. Version control. Post-deployment monitoring. Clear documentation. Bias assessment. Human oversight. A way to challenge outputs. A way to stop using the tool when performance drifts.
This may sound technical.
It is not.
It is patient safety.
If an oncology drug changes, regulators pay attention. If an AI model changes, cancer systems need to show the same seriousness.
Clinical trial matching may be the test case
Clinical trial matching is one of the clearest areas where AI could make a practical difference.
Too many patients never hear about trials for which they may be eligible. Trial criteria are complex. Clinicians are overloaded. Trial slots change. Referral networks are uneven. Patients outside major centres are often structurally excluded from research and innovation.
AI could help identify options faster, read records more efficiently and bring trial opportunities into the moment when treatment decisions are actually being made.
That could be transformative.
But only if the system is transparent.
A patient and clinician need to know why a trial was suggested, which criteria were met, which were uncertain and which data were missing. Without that, AI trial matching risks becoming another black box in a system patients already struggle to navigate.
The goal should not be to make trial matching look efficient.
The goal should be to make access to research fairer.
Patients are already using AI
The politics of AI in oncology is not only institutional.
Patients are using AI too.
They are asking chatbots to explain scan results, symptoms, treatment options, laboratory findings and side effects. Some will come to appointments with AI-generated summaries before they have had enough time with a clinician.
Cancer systems can dismiss this, or they can prepare for it.
The wrong response is to tell patients not to use AI while hospitals, insurers and health systems quietly use AI themselves.
The better response is transparency and literacy.
Patients should know when AI is being used in their care. Clinicians should be supported to discuss AI-generated information clearly. Health systems should help people understand what AI can and cannot do.
Trust will not come from pretending AI is invisible.
It will come from making its role visible.
The real question after ASCO
ASCO 2026 showed that AI is becoming part of oncology’s future.
But the most important question is not whether AI can do more.
It is whether cancer systems know when AI should be trusted.
That requires more than excitement from innovators and adoption by hospitals. It requires political choices about evidence, liability, procurement, data governance, patient consent, professional responsibility, reimbursement and equity.
AI may help oncology become faster, smarter and more personalised.
But only if it is governed as carefully as the treatments it supports.
Because in cancer care, a powerful tool is not enough.
It has to be accountable.
