Cisco unveiled a major open source initiative at the RSA Conference today, announcing the release of an AI model specifically trained for cybersecurity applications. In a keynote address at RSA Conference 2025 in San Francisco, Jeetu Patel, EVP and Chief Product Officer, described how Cisco’s new foundation model, trained on a carefully selected subset of cybersecurity data, is designed to run efficiently on smaller hardware footprints compared to general-purpose AI models. Patel emphasized that the open sourcing of the model and its tooling is intended to help the broader security community defend against increasingly sophisticated, machine-scale cyberattacks.
Patel outlined two priorities for AI in security: securing AI itself from adversarial threats, and using AI to strengthen cybersecurity operations at machine scale. The open source AI model will enable developers and security teams to build smarter agents that can automatically investigate threats, validate alerts, and recommend containment actions—dramatically improving security efficacy while reducing operational costs. Cisco stressed that in a world where attacks are being automated, human-scale defenses are no longer sufficient, and domain-specific AI expertise will be critical.
- Cisco is open-sourcing a cybersecurity-specific AI foundation model and related tooling via Hugging Face.
- The model is trained on 5 billion tokens selected from a curated cybersecurity corpus, out of a larger 900 billion token dataset.
- It is highly efficient, capable of running on just one or two NVIDIA A100 GPUs, compared to much larger hardware requirements of generic LLMs.
- The model enables automated security workflows, including alert validation, threat investigation, and action recommendations.
- Cisco advocates for securing AI models themselves against external manipulation and hallucination, alongside using AI to bolster cyber defenses.
- Patel stated that domain-specific AI for cybersecurity is necessary, contrasting it with using generalized AI models built for broader tasks.
- Cisco’s goal is to foster an open ecosystem where security researchers can collaborate against adversaries rather than compete over proprietary models.
- The initiative is part of a broader Cisco strategy to advance “super-intelligent security” through orchestrated AI agents.
“The true enemy is not our competitor—it’s the adversary. By open sourcing our cybersecurity AI model, we can collectively raise the bar to keep our organizations, our nations, and humanity safe,” said Jeetu Patel, EVP and Chief Product Officer, Cisco.
