Cisco’s latest innovation in cybersecurity, the Antares AI model, promises to revolutionize how vulnerabilities in source code are detected. Unveiled by Cisco Foundation AI, the Antares model is an economical yet effective solution for organizations seeking to enhance code security. This small language model (SLM) is now available on Hugging Face, demonstrating Cisco’s commitment to advancing technology in a cost-effective manner.
Understanding the Antares Model
The Antares AI model series comprises two versions: Antares-350M and Antares-1B. These open-weight models are designed to address one of the most challenging aspects of cybersecurity—identifying known vulnerabilities within a codebase. By providing a targeted solution, Cisco aims to aid security teams in swiftly locating and addressing potential threats in their software systems.
Unlike larger, closed AI models that can be costly and subject to data sovereignty issues, the Antares models offer a more affordable and flexible alternative. This makes them particularly appealing to organizations dealing with stringent data regulations, as the models enable in-house processing without requiring external data transfer.
Efficiency and Accuracy in Vulnerability Detection
One major advantage of the Antares model is its ability to diminish the occurrence of false positives, a common issue with smaller, open-weight general language models (GLMs). Cisco’s design leverages a search process akin to a human investigator, systematically examining code patterns and narrowing down on files likely to contain vulnerabilities. This ensures both accuracy and efficiency, saving valuable time and resources for security teams.
Targeted at budget-conscious entities such as universities and public-sector teams, Antares is engineered to provide high-caliber security analysis without the hefty price tag typically associated with large-scale AI solutions. By integrating AI in initial stages of vulnerability triage, organizations can enhance their security posture significantly.
Cisco’s Commitment to Security Innovation
To validate the effectiveness of Antares, Cisco has introduced the Vulnerability Localization Benchmark, a 500-entry test designed to challenge models in navigating new codebases and identifying vulnerabilities. When tested against leading models like OpenAI’s GPT-5.5 and Z.ai’s GLM-5.2, Antares proved to be both faster and more cost-effective.
Amin Saberi, a Stanford University professor, highlighted Antares’ potential to democratize advanced security tools, making them accessible to teams with limited resources. This aligns with Cisco’s broader vision of developing AI tools that facilitate everyday security operations, regardless of an organization’s size or budget.
Cisco’s foray into AI-assisted security reflects its dedication to fostering a robust foundation for digital defense. Through open specifications and reusable security knowledge, Cisco aims to empower all security practitioners to leverage AI effectively, ensuring comprehensive protection against emerging threats.
