This Phase 1/2 study evaluates the safety, tolerability, and preliminary anti-tumor activity of an allogeneic dual-target chimeric antigen receptor natural killer (CAR-NK) cell product in adults with advanced or metastatic colorectal cancer (CRC). Participants are assigned to one of three dual-target arms based on tumor antigen co-expression: (1) CEA+GUCY2C, (2) CEA+HER2, or (3) GUCY2C+HER2. Following dose escalation, the most suitable target pair (based on safety, feasibility, and early efficacy/biomarker signals) will be selected for dose expansion.

Colorectal cancer remains a major cause of cancer-related morbidity and mortality. CAR-NK therapy is being explored as a potentially safer, 'off-the-shelf' adoptive cell therapy platform with lower risk of graft-versus-host disease and typically less severe cytokine release than CAR-T in early experiences.

Target selection for solid tumors is limited by tumor heterogeneity and on-target/off-tumor expression. To reduce antigen escape and increase tumor selectivity, this study tests dual-target recognition against CRC-associated antigens: CEA (CEACAM5) and GUCY2C (GCC), with a HER2/ERBB2-positive subset as an exploratory population. GUCY2C is frequently retained in primary and metastatic CRC and is a well-established CRC target antigen. HER2 amplification/overexpression occurs in a minority of metastatic CRC and can be identified by validated IHC/ISH or genomic assays. Biomarker assessment and target-pair selection: Tumor tissue (archival or fresh) is tested by central laboratory immunohistochemistry (IHC) for CEA and GUCY2C, and by HER2 testing per colorectal cancer criteria (IHC with reflex amplification testing as needed). Participants must meet co-expression thresholds for one antigen pair and are assigned accordingly. After Part A (dose escalation), a prespecified selection algorithm will nominate the target pair for expansion using a composite score including DLT rate, feasibility/manufacturing success, ORR/DCR signals, CAR-NK persistence, and evidence of target engagement. Treatment plan: Participants receive lymphodepleting chemotherapy followed by infusion of the assigned dual-target CAR-NK cells. A second infusion in the same cycle may be permitted in selected dose levels if safety criteria are met. Hospitalization/close monitoring is required during the early post-infusion period.

Inclusion Criteria:

• Histologically confirmed colorectal adenocarcinoma that is unresectable or metastatic and has progressed after, is intolerant to, or is ineligible for standard therapies.
• Measurable disease per RECIST v1.1 (unless in minimal residual disease (MRD) or post-resection cohorts if a future amendment is planned).
• Tumor antigen co-expression meeting central lab thresholds for one of the following pairs: CEA+GUCY2C, CEA+HER2, or GUCY2C+HER2.
• ECOG performance status 0-1.
• Adequate organ function (hematologic, renal, hepatic, and cardiac) as defined in protocol.
• Recovered to Grade <=1 from prior therapy-related toxicities (except stable Grade 2 neuropathy or alopecia).
• Life expectancy >= 12 weeks.
• Willingness to use effective contraception during study and for a protocol-defined period after cell infusion.

Exclusion Criteria:

• Active, uncontrolled infection (including uncontrolled HBV/HCV) or known uncontrolled HIV infection.
• Active CNS metastases that are symptomatic or require escalating steroids. (Stable treated CNS disease may be allowed per protocol.)
• Prior gene-modified cellular therapy (CAR-T/CAR-NK/TCR-T) within 6 months, or any prior therapy that in the investigator's judgment increases risk of severe toxicity.
• Clinically significant autoimmune disease requiring systemic immunosuppression within the past 6 months.
• Concurrent anti-cancer therapy (other than protocol-permitted bridging) during the DLT window.
• Pregnant or breastfeeding.
• Significant cardiovascular disease (e.g., recent MI, uncontrolled arrhythmia), uncontrolled pulmonary disease, or other severe comorbidity that would increase risk.
• Known hypersensitivity to study chemotherapy components (fludarabine/cyclophosphamide) or required supportive medications.
• Any condition that, in the investigator's opinion, would interfere with study participation, safety monitoring, or interpretation of results.

Open-label administration; endpoint assessment follows RECIST v1.1 by investigator and/or independent review (if used).