AURORA Puts AI Co-Scientists in Cancer Grand Challenges Labs

Science has a bottleneck that has little to do with lab benches: there is simply too much published evidence and too much data for any research team to read, compare and connect by hand. On November 6, 2026, Cancer Research UK announced that three of its Cancer Grand Challenges teams will begin a rigorous test of whether AI “co-scientists” can help close that gap.

The test is the next stage of Project AURORA, an initiative the charity describes as an £8 million project. A team at Harvard University developed the specialized co-scientist platform with Cancer Research UK funding, and the Cancer Grand Challenges teams will now use it in live research and evaluate how well it works.

What a “co-scientist” is meant to be

AURORA’s co-scientists are designed to act as virtual lab assistants. Cancer Research UK stresses what they are not: they cannot make scientific decisions on their own. Guided by the expertise of real researchers, they analyze large amounts of data to surface clues that might point to new directions.

Two design choices stand out. First, the system is built to stay tied to evidence. Cancer Research UK says the co-scientists always draw on properly reviewed scientific resources and research data, and that when there is not enough evidence to support a conclusion, they are designed to say so. That is a direct response to the habit of general chatbots to produce confident but invented answers.

Second, AURORA uses agentic AI. In plain terms, the co-scientists can guide themselves through a task step by step, reasoning about what to do next. Researchers keep control and oversight but do not have to sit at the keyboard all day, since they can ask the co-scientist to comb through large libraries and databases without constant check-ins.

Three teams, three problems

The charity says the technology will be applied to three separate problems, each owned by a Cancer Grand Challenges team:

• NexTGen will explore how immune cells can be modified to better find and break down solid tumors in children and young people, including which features of cancer cells to target and how to overcome obstacles that stop immune cell treatments from working.
• PROMINENT will examine why some cells carrying cancer-linked genetic changes stay healthy while others turn cancerous, using the co-scientists to suggest explanations to test, including the possible influence of age, sex and exposure to cancer-causing substances.
• MATCHMAKERS will study how T cells recognize cancer, testing which markers on cancer cells make the most promising targets and which immune cells are best equipped to attack them, work that could help shape more personalized treatments.

Humans still run the experiments

The announcement is careful about the limits. An AURORA co-scientist could, for example, spot a hidden pattern in a large database and propose a new hypothesis, but human scientists must still test it and evaluate the results before drawing conclusions. The trial also serves as an evaluation: the teams will assess AURORA’s outputs, and their feedback will be used to improve the system so it can make better suggestions for what to investigate next.

Marinka Zitnik, an associate professor at Harvard Medical School and the lead researcher behind AURORA, framed the aim as building a trusted, rigorously evaluated tool by testing it against real scientific questions and experimental evidence. Professor Charles Swanton, who heads the Cancer Grand Challenges Scientific Committee, described cancer as an “immense landscape of biological complexity” that no team can map alone.

What it means for research teams

The interesting part of AURORA is less the model than the method. Putting an AI system inside funded, real-world research programs, and judging it by whether it improves the research process, is a more demanding test than a benchmark score. If co-scientists can bring together relevant evidence faster, widen the range of explanations a team considers and reveal links that are hard to see, they could make exploration more efficient.

Cancer Research UK is realistic about expectations. It says AI is unlikely to cure cancer on its own, and that some of the experiments a co-scientist recommends will not lead to discoveries, just as many human-chosen experiments do not. Testing ideas that turn out to be wrong is part of how science builds understanding, and the charity suggests that helping researchers get things wrong a little faster could still help the field move forward.

For research leaders outside oncology, the practical lesson is about governance. The AURORA design puts three safeguards in the same package: answers tied to reviewed sources, an explicit admission when evidence is thin, and a human who owns every scientific decision. Teams evaluating their own AI assistants could borrow that checklist, then ask the same question Cancer Research UK is asking here: does the tool measurably help the work, or does it only sound impressive?

This article reports on research tooling and is news only. It is not medical advice, and AURORA does not diagnose or treat patients.

The bottom line

AURORA turns the “AI for science” idea into a field test with named teams, named problems and a plan to measure results. Whatever the outcome, the evidence it produces about when agentic co-scientists help, and when they do not, will be useful to any research organization weighing similar tools.

Sources

• Can AI “co-scientists” help answer cancer’s biggest questions? (Cancer Research UK, Nov. 6, 2026)
• Cancer Grand Challenges: home page