Cisco has introduced Cisco Antares, a new family of compact AI models built to detect software vulnerabilities. The announcement was made in Dubai, United Arab Emirates, on July 28, 2026.

Cisco Antares models are designed to run entirely on an organization’s own systems. As a result, companies do not need to send sensitive source code to the cloud. This approach helps reduce cost while keeping proprietary data private.

Finding vulnerabilities in software is a difficult and time-consuming task. However, general-purpose AI tools often require cloud access, which raises privacy concerns. Consequently, many organizations with strict compliance rules have struggled to adopt AI-based security scanning.

To address this gap, Cisco built Cisco Antares as a family of small language models, or SLMs. These models are lightweight, efficient, and purpose-built for cybersecurity tasks. Moreover, they operate locally within a company’s secure environment.

Cisco is releasing two versions of the technology, Antares-350M and Antares-1B, openly to developers and the wider security community. Therefore, smaller teams with limited budgets can now access advanced AI security tools that were previously out of reach.

Fady Younes, Managing Director for Cybersecurity at Cisco Middle East, Türkiye, Africa, Caucasus and Central Asia, commented on the launch. He said regional organizations are accelerating their digital capabilities, so securing software without compromising data privacy is critical. He added that Antares gives security teams local AI power, allowing them to pinpoint vulnerabilities faster while keeping source code within their own environment.

The Antares launch includes several key features. First, the models support privacy-first security, since local processing removes the need to send code to external servers. This makes the technology especially useful for the public sector, universities, and organizations with strict data sovereignty rules.

Second, benchmark testing shows that Antares outperforms many larger, more expensive AI models in critical security tasks. At the same time, it operates at a fraction of the cost.

Third, Antares mimics human-like investigation. Instead of relying on rigid rules, it reads vulnerability descriptions, searches for relevant code, and changes direction when a path proves unhelpful. Eventually, it narrows down the file paths most likely to contain a threat.

Finally, Cisco stated that the open release supports its broader goal of democratizing AI security. In particular, smaller teams that previously lacked access, budget, or resources can now deploy AI-assisted vulnerability detection.

Overall, Cisco is moving beyond model development alone. Instead, the company aims to build the ecosystem and standards needed for practical, trustworthy AI adoption in the enterprise. With Cisco Antares, Cisco continues to expand its role in shaping secure, privacy-focused AI tools for the cybersecurity industry.

Related post

View all