tech
Cisco bets on small AI for cybersecurity
The company is open-sourcing two small-language models designed just to hunt for software bugs.
TL;DR
- Cisco has launched two small, open-weight AI models called Antares to detect software vulnerabilities.
- The Antares models performed comparably to larger models like GPT-5.5 and GLM-5.2 on a new Cisco benchmark.
- Antares-350M and Antares-1B will be available on Hugging Face to verified cyber defenders.
- These smaller models are designed to run locally, allowing companies to keep sensitive source code private.
- Cisco's testing showed Antares scanning 500 repositories for under $1 in about 15 minutes, versus hours and over $100 for GPT-5.5.
- Antares is trained to act as a security investigator, searching for vulnerabilities rather than assisting with general coding.
- Cisco collaborated with U.S. government agencies on the models' safety and release, and will vet users to prevent access by adversaries.
- Cisco is exploring an industry consortium to further develop open AI security tools.