{"resourceId":"cruk-aurora-ai-co-scientists-cancer-grand-challenges-20261106","versions":[{"version":"external-c626d753a38ac80e94a56d83198f425dc15a5f723ab2828d64630558aee2bcfe","resource":{"id":"cruk-aurora-ai-co-scientists-cancer-grand-challenges-20261106","title":"Can AI \"co-scientists\" help answer cancer's biggest questions?","organization":"Cancer Research UK (covering Harvard-led AURORA / Cancer Grand Challenges)","sector":"Biomedical research funding / university cancer research programmes","geography":"United Kingdom / international Cancer Grand Challenges (Harvard platform; UCL teams among participants)","publishedAt":"6 November 2026 (Cancer Research UK Cancer News)","publicationDate":"2026-11-06","eventDate":"2026-11-06","sourceName":"Cancer Research UK Cancer News","sourceLabel":"Gunn, T., \"Can AI co-scientists help answer cancer's biggest questions?,\" Cancer Research UK Cancer News, 6 Nov 2026","sourceUrl":"https://news.cancerresearchuk.org/2026/11/06/can-ai-co-scientists-help-answer-cancers-biggest-questions/","evidenceClass":"independent-reporting","outcomeClass":"emerging","topics":["developers-agents","knowledge-work","governance-procurement"],"finding":"From 6 November 2026, Cancer Grand Challenges teams begin a rigorous test of agentic AI 'co-scientists' from the £8m AURORA project, developed at Harvard with Cancer Research UK funding. Teams PROMINENT, MATCHMAKERS and NexTGen will use the tools inside ongoing cancer programmes to analyse evidence, generate ideas and suggest experiments. CRUK states the co-scientists cannot make scientific decisions on their own and remain guided by real scientists' expertise, with researchers retaining control and oversight while agents carry longer tool-using tasks. This is an evaluation kickoff narrative, not reported outcome metrics from the new phase.","sledRelevance":"Shows how major biomedical research programmes are institutionalizing AI research agents with explicit human decision rights—useful precedent for university research administration and cancer institutes.","evidence":"Funder news article describing programme design and governance intent; corroborated in related Cancer Grand Challenges / UCL notices; no measured discovery acceleration yet from this phase.","architectureImplications":"Agentic assistants with scientific tool/data connectors (ToolUniverse lineage noted in CGC materials) behind human approval for experimental commitments.","governanceImplications":"Encode human-final decision rights, uncertainty handling, and evaluation plans before campus-wide co-scientist rollouts.","securityPrivacyImplications":"Tool allowlists and data-access agreements needed for cancer datasets; article does not specify technical controls.","caveats":"Launch/evaluation announcement without results; funder communication; clinical effectiveness not claimed.","streamIds":["research"],"roles":{"sales":"Deployment framing: Cancer Research UK reports that from 6 Nov 2026, Cancer Grand Challenges teams (PROMINENT, MATCHMAKERS, NexTGen) begin rigorously testing Harvard-built AURORA agentic 'co-scientists' inside live cancer programmes under an £8m project—explicitly keeping scientific decisions with humans. Relevant to university cancer institutes evaluating research agents. Sell governed co-pilot deployments with evaluation hooks, not autonomous clinical or experimental decision rights.","engineering":"Architecture framing: agentic co-scientists connected via ToolUniverse-style tool and data bridges to literature and programme repositories, with human final authority on hypotheses and experiments. Log agent suggestions versus experimenter choices so later outcome evaluation is possible. Separate research-assist agents from any clinical decision support path, and enforce least-privilege allowlists on tools that can touch sensitive cancer datasets. Surface uncertainty explicitly when evidence is thin rather than forcing a ranked experiment list.","delivery":"Delivery framing: onboard one challenge team with clear decision rights, immutable audit logs, and a pre-registered evaluation of whether AI-suggested experiments improve progress metrics versus baseline triage. Scientific leads own decisions; informatics owns tool connectors and access agreements. Acceptance is measured experimental follow-through and documented human overrides—not chatbot satisfaction scores. Publish the evaluation plan before scaling beyond the pilot team."},"roleDetails":{"sales":{"customerProblem":"Cancer research teams face overwhelming literature/data volume but cannot safely delegate scientific decisions to chatbots.","stakeholders":["Cancer institute director","Challenge team PIs","Research computing / bioinformatics","Research integrity office"],"discoveryQuestions":["Which decisions stay human-only under your AI research policy?","How will you score co-scientist suggestions against later wet-lab outcomes?","What tools/data will agents be allowed to call?","How is uncertainty communicated when evidence is weak?"],"valueHypothesis":"Governed agentic co-scientists may accelerate hypothesis triage if humans retain decision rights and programmes evaluate suggestions against experiments—as AURORA is designed to test.","potentialEngagement":"A single-team governed pilot with ToolUniverse-like connectors, decision logging, and a 90-day evaluation plan.","unsupportedClaims":["Proven acceleration of cancer discovery outcomes (trial is starting)","Safe autonomous experimental decision-making","Clinical care recommendations"]},"engineering":{"fit":"Fit for university cancer/biomedical programmes evaluating research agents; not a clinical product architecture.","architectureIntegration":"Agent runtime + scientific tool/data adapters + audit log + human approval UI before experiment prioritization commits.","prerequisites":["Programme data access agreements","Tool allowlists","Human decision policy","Evaluation metrics"],"constraints":["Funded evaluation just starting 6 Nov","Funder/news description not a methods paper","UK/international team context"],"security":"Least-privilege tool access; no PHI/clinical actuation paths; retain immutable suggestion/decision logs.","proofOfValue":["Fraction of AI-suggested experiments run","Outcome hit rate vs baseline triage","Documented human overrides"]},"delivery":{"work":"Policy + allowlist; Connect tools; Train team on oversight; Run evaluation protocol","dependencies":["PI scientific ownership","Infomatics connectors","CRUK/CGC programme rules where applicable"],"ownership":"Challenge PI for science; institute informatics for platform.","skillsAdoption":"Researchers learn to task agents and critique outputs; staff monitor audit trails.","governanceCheckpoints":["Human decision attestation","Tool-allowlist changes reviewed","Quarterly evaluation readout"],"acceptanceCriteria":["Proposed: 100% of experimental commitments human-attested","Proposed: logged suggestion→decision→result chain","Proposed: published evaluation plan"],"risks":["Automation bias","Scope creep into clinical advice","Overclaiming before AURORA results exist"]}},"retrievedAt":"2026-11-06T21:58:56-06:00","enrichedAt":"2026-11-06T22:03:30-06:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Positions AI as labor for literature/data triage while preserving scientist judgment roles; training needs for oversight not detailed.","procurementImplications":"Research-agent purchases for biomedical institutes should include evaluation protocols and contractual human-decision requirements.","operatingModelImplications":"Co-scientist-as-assistant model with PIs owning scientific decisions and informatics owning connectors/logs.","peerReview":"unknown","sourceVerification":{"openedUrl":"https://news.cancerresearchuk.org/2026/11/06/can-ai-co-scientists-help-answer-cancers-biggest-questions/","referenceExcerpt":"They can’t make any scientific decisions on their own, but, guided by real scientists’ expertise, they can analyse vast amounts of data","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}