Skip to content
CERT-EU·Indexed by this site · Originally published· Tools →· Its place in the scan record →

CERT-EU open-sources Vigilo, its internally developed AI code security review tool

CERT-EU open-sourced its internally developed harness for code review

AI summary

CERT-EU has open-sourced Vigilo, a code security review tool it developed internally, to guide AI agents in reviewing code from a cybersecurity perspective. Vigilo orchestrates multiple agents, each focused on one class of vulnerability (such as injection, authentication, authorization, cryptography, and supply chain dependencies), which exchange findings through a shared knowledge base, and supports two modes: white-box (analyzing source code and optionally a running instance) and gray-box (testing only a running application, simulating an external attacker holding valid credentials). Each finding goes through an adversarial review step that re-examines severity and merges duplicates; when source code is available, the tool can also automatically generate a patch for each confirmed finding, placed on a separate branch for code maintainers to review. Vigilo took part in an ENISA proof of concept for source code testing using frontier AI models, and can now be deployed as a command-line tool or a multi-user web platform on an organization's own infrastructure, supporting cloud or local models; CERT-EU cautions that findings and patches proposed by AI-assisted review should be verified by the code owner before fixes are applied.

Read originalcert.europa.eu