Cisco Introduces Antares: a family of super-efficient AI models for detecting software vulnerabilities

Compact models, released openly by Cisco, run entirely on organizations’ on-prem systems, identifying security flaws at reduced cost while keeping sensitive data private

Today Cisco has introduced Antares, a family of small language AI models (SLMs) purpose-built to tackle one of the most difficult and time-consuming challenges in cybersecurity: finding hidden security flaws within software codebases, on-premises, without externalizing organizations’ data

As organizations race to secure their digital infrastructure, finding vulnerabilities quickly is critical. However, using general-purpose AI to scan code can be expensive and often requires sending sensitive, proprietary source code to the cloud, representing a major hurdle for organizations with strict privacy and compliance requirements

To help address this, Cisco has introduced Antares. These models are compact, cost-effective, and capable of running locally within an organization’s own secure environment

“As regional organizations accelerate their digital capabilities, securing complex software without compromising data privacy is critical and urgent,” said Fady Younes, Managing Director for Cybersecurity at Cisco Middle East, Türkiye, Africa, Caucasus and Central Asia (METAC). “With Antares, we are giving security teams the power of AI locally so they can pinpoint vulnerabilities faster while keeping sensitive source code firmly within their own secure environment”

Cisco is releasing two of these models (Antares-350M and Antares-1B) entirely openly to the broader developer and security community

Key highlights of the Antares launch include

            Privacy-First Security: Because the models are compact enough to run locally, security teams do not need to send sensitive source code to the cloud. This makes Antares ideal for the public sector, universities, and organizations with strict data sovereignty rules

            Unmatched Efficiency: Benchmark testing shows that the Antares models outperform many larger, more expensive AI models in critical security tasks, operating at a fraction of the cost

            Human-Like Investigation: Rather than relying on rigid rules, Antares reads vulnerability descriptions, searches for relevant code, changes direction if a path is unhelpful, and narrows down the file paths most likely to contain a threat

            Democratizing AI Security: By making these tools openly available, Cisco is unlocking the power of AI-assisted security for smaller teams that previously lacked the access, budget or resources to deploy proprietary large language AI models

With the release of Antares, Cisco is moving beyond simply building models; it is helping create the ecosystem and standards needed for practical, trustworthy enterprise AI adoption