Cisco Launches Antares: Open-Weight AI Models for Practical Vulnerability Detection
Key Takeaways
- ▸Cisco releases Antares-350M and Antares-1B as open-weight security small language models specifically designed to locate known vulnerabilities in codebases
- ▸These compact models outperform larger closed and open-weight models while running locally, keeping sensitive source code on-premises and reducing inference costs
- ▸The open-weight approach democratizes AI-powered security tools for universities, public sector institutions, and smaller organizations previously unable to afford frontier-scale AI security solutions
Summary
Cisco introduced Antares, a family of security small language models (SLMs) purpose-built for vulnerability localization—identifying where known vulnerabilities exist within codebases. The company is releasing two models, Antares-350M and Antares-1B, as open-weight models available on Hugging Face. Benchmark testing shows these compact models outperform many larger closed and open-weight models in vulnerability detection at a fraction of the cost.
A key advantage of Antares is its efficiency and locality: the models are compact enough to run on-premises or locally, eliminating the need to send sensitive codebases to the cloud. This addresses a critical security and privacy concern for enterprises while reducing inference costs. The small model size makes AI-assisted security practical for universities, public sector institutions, and smaller security teams that previously lacked resources for token-intensive AI models.
Beyond the models themselves, Cisco is releasing Foundry Security Spec (an open specification for security AI tools), CodeGuard (secure coding guidance), and a new Vulnerability Localization Benchmark. Together, these components aim to establish practical, trustworthy AI tools and standards for enterprise adoption of security AI.
- Cisco is establishing an ecosystem with Foundry Security Spec, CodeGuard, and a new benchmark to support practical, trustworthy enterprise AI security adoption
Editorial Opinion
This is a significant step toward practical AI security. By open-sourcing efficient models that run locally, Cisco addresses enterprise security's core concern: advanced threat detection without sacrificing code privacy. The strategy of pairing open models with open standards (Foundry Security Spec) and clear guidance (CodeGuard) could set a template for how AI vendors approach security—where trust and practicality matter as much as raw capability. This democratization of AI security is likely to accelerate adoption across smaller organizations and institutions that were previously excluded from frontier AI tooling.
