Cisco Builds Cheaper AI Tools to Hunt Security Flaws in Software Code
The company's new Antares models scan source code for known vulnerabilities at a fraction of what larger AI systems cost. Here is what that means for businesses that write or buy software.
Key points
- Cisco released a family of AI models called Antares in 2025, designed specifically to find security weaknesses in software source code.
- Antares is described as "open-weight", meaning other developers can inspect and build on the underlying model.
- The models cost considerably less to run than comparable large AI systems.
- Their primary job is to identify known vulnerabilities, meaning software flaws that security researchers have already catalogued and named.
Cisco has released a set of AI models aimed at one specific problem: finding security flaws buried inside software code before criminals can exploit them. The product family is called Antares, and SecurityWeek first reported its launch.
Every piece of software is built from source code, the human-readable instructions a developer writes. Hidden inside that code can be vulnerabilities, meaning weaknesses an attacker could use to break in or seize control of a system. Spotting those weaknesses before shipping is one of the hardest jobs in the industry. Antares is built to do it automatically.
The models are "open-weight": the underlying structure is published so other companies can inspect, adapt and plug them into their own tools. That's different from a closed system where the inner workings stay secret. Capital One took a similar open approach last month when it released VulnHunter free of charge, arguing no single company can solve this problem alone.
The cost angle matters most. Running large AI models is expensive, billed by computing hour. Cisco says Antares delivers comparable results at a fraction of that cost, which could make automated code-security checks realistic for smaller teams that currently skip them because the price is prohibitive.
Antares focuses on known vulnerabilities, flaws already discovered and catalogued under a CVE identifier (Common Vulnerabilities and Exposures, the standard numbering system the industry uses). Developers sometimes accidentally reuse patterns already proven dangerous. An AI model that flags those patterns quickly and cheaply could stop problems before a product ships. Given that eight in ten organisations carry a backlog of unresolved flaws stretching back more than a year, tools that lower the cost of routine scanning matter more than the launch noise around them.
Should you worry if you don't write software?
Not directly today. More teams running better automated checks means fewer products reach the public with easily preventable holes. That benefits anyone who uses software.
If your company develops software in-house, it's worth asking your technical team whether automated code-scanning is part of the build process. It doesn't have to be expensive any more.
