Cisco Introduces Antares AI Models for Code Vulnerability Detection

Overview of Antares AI Models

Cisco Foundation AI has unveiled Antares, a new family of small language models specifically created to identify vulnerabilities in source code. Named after a prominent star in the Scorpius constellation, Antares includes two models: Antares-350M and Antares-1B. Both models are offered as open-weight on Hugging Face.

Targeting Vulnerabilities in Codebases

Antares is designed to help security teams pinpoint existing vulnerabilities in their codebase, a traditionally challenging problem. The model aims to replicate the investigative approach a human expert would take, processing vulnerability descriptions, relevant code patterns, and narrowing down the search based on evidence found. This method improves accuracy and reduces the likelihood of false positives.

Cost-Effectiveness and Data Sovereignty

Current market solutions for identifying vulnerabilities include expensive closed large language models (LLMs) and open-weight general language models (GLMs). While closed models lower false positives, their high costs can deter organizations, especially in cases where code must stay within regulatory frameworks. Antares combines low operational costs with the benefits of managing data internally, addressing these concerns.

Market Implications

The introduction of Antares could reshape how companies approach vulnerability detection within codebases. By offering a competitive alternative to high-cost models while maintaining a level of accuracy, Cisco may appeal to a range of businesses needing effective security solutions. Organizations can benefit from reduced costs and enhanced data governance in their vulnerability assessments.