Background: Anastomotic leaks (AL) remain a dreaded complication after colorectal surgery, and is associated with increased morbidity, mortality and length of hospital stay. This systematic review examines the potential of artificial intelligence (AI) to predict AL. Methods: Medline/Pubmed, Cochrane, EMBASE, Web of Science, and Google Scholar were searched for studies of any design evaluating AI models for predicting AL in patients undergoing colon or rectal surgery for any indication. Subgroup analyses were conducted on pre and intra-operative data as well as rectal cancer. Results: Fourteen studies (26,932 patients; mean age 66.7 years; 57% male) were included. Most patients (87%) had surgery for cancer; 31.1% had rectal surgery. AI models, including support vector machines, LASSO-logistic regression, and neural networks, showed promising predictive accuracy, with AUC values often exceeding 0.80. Models using exclusively pre- and intra-operative data (7 studies) yielded AUCs between 0.73 to 0.89. In studies focusing on rectal cancer (4 studies), AI models achieved AUCs ranging from 0.78 to 0.88. Traditional risk factors like low tumour location, longer operative time, and higher ASA scores were consistently identified and included in many models. Novel methods such as incorporating free clinical text data with laboratory values (AUC of 0.97) or real-time Indocyanine green angiography (AUC 0.842) were reported. However, studies exhibited considerable heterogeneity in methodology and reporting. Conclusion: AI-based models hold significant potential in predicting AL after colorectal surgery to facilitate personalised risk assessment and guide clinical practice.