GitHub announced About adding to service Code Scanning experimental Machine training systems To identify common types of peptics in the code. At the test stage, the new functionality is still available for repositories with the code in JavaScript and TypeScript. It is noted that the use of machine learning system made it possible to significantly expand the spectrum of detectable problems, when analyzing which the system is now not limited to testing typical templates. In particular, the new system allows you to find errors that lead to cross-site scripting (XSS), distorting file pathways (for example, via “/ ..”), substitution SQL- and NOSQL queries.
Code Scanning service allows you to identify vulnerabilities at an early stage of development through the scanning of each Git Push operation for potential problems. The result is attached directly to the Pull request. Previously, the test was carried out using the codeql , analyzing templates with typical examples of a vulnerable code (CodeQL allows you to form a vulnerable code template to detect the presence of such vulnerability in the code other projects). A new engine that uses machine learning can define not previously known vulnerabilities as it is not tied to a code of code templates describing specific vulnerabilities. The price of such an opportunity is to increase the number of false positives compared to CodeQL-based inspections.