Cisco Foundation AI

Anyone who has worked an incident knows the real story never fits in a small prompt window. In this new article, Dhruv Kedia breaks down Foundation-sec-1.1-8B-Instruct — Cisco Foundation AI’s latest long-context model designed for true end-to-end security reasoning. With a 64K-token window, it’s built for full investigations, full policy reviews, and full context — the way analysts work. Follow us on LinkedIn for more AI + security insights Download our models on Hugging Face: https://lnkd.in/gCHP5cxN Explore the Foundation AI Cookbook: https://lnkd.in/g49p2ctb

The November inaugural edition of the Cisco Foundation AI Monthly Brief is here. This first issue brings together highlights from across Foundation AI — from advances in model infrastructure, structured outputs, and AI provenance to the launch of our new AI for Humans series, designed to make AI more approachable and practical. Explore the full collection below, with direct links to each original post and publication.

Each post in our AI for Humans series explores a new dimension of how artificial intelligence is transforming cybersecurity — from its origins to its most advanced forms. In this installment, Vandita Anand traces AI’s remarkable evolution from the early rule-based expert systems of the 1980s to today’s agentic AI — systems that can plan, act, and adapt in real time. It’s a thoughtful reflection on how far we’ve come, how fast we’re moving, and why understanding this history is essential to shaping a secure AI-powered future. Follow us on LinkedIn for more AI + security insights Download our models on Hugging Face: https://lnkd.in/gCHP5cxN Explore the Foundation AI Cookbook: https://lnkd.in/g49p2ctb

As Cybersecurity Awareness Month comes to a close, we have a very special contribution to share. Shailaja K. Shankar, SVP of Cisco's Security Business Group Engineering, reflects on one of the most important shifts of our time - how AI is transforming not only how we build, but how attackers think. In her piece, Building Fast, Building Smart: Cybersecurity in the Age of AI and Agentic Development, Shaila explores the dual edge of AI-powered software creation - its ability to accelerate innovation, and the parallel rise of new security risks like the "synthetic monoculture" of AI-generated code. Shaila reminds us that in this new era, developers are no longer just coders - they are curators of trust. This is a must-read perspective from one of Cisco's foremost technology leaders on how to build fast, build smart, and build securely. Follow us here on LinkedIn for more AI + security insights Download our models on Hugging Face: https://lnkd.in/gCHP5cxN Explore the Foundation AI Cookbook: https://lnkd.in/g49p2ctb

As AI systems become more powerful and interconnected, the question of where models come from — and how they’re built — is becoming as important as what they can do. Across governments and industries, model provenance is emerging as a cornerstone of AI trust and compliance. It connects technical transparency with policy accountability, helping organizations navigate a world where innovation, regulation, and geopolitics increasingly intersect. In “The Importance of AI Provenance through a Policy Lens,” Alie Fordyce from Cisco Foundation AI explores how provenance has evolved from a technical safeguard into a policy requirement — one that shapes how AI is governed, deployed, and trusted across global ecosystems. Follow us here on LinkedIn for more AI + security insights Download our models on Hugging Face: https://lnkd.in/gCHP5cxN Explore the Foundation AI Cookbook: https://lnkd.in/g49p2ctb

The closer AI gets to real security workflows, the more structure starts to matter. Free-form text is fine for reading — but automation, integration, and auditability all depend on clean, predictable outputs. At Cisco Foundation AI, we’ve been exploring how large models can move beyond words to deliver consistent, schema-compliant JSON that real systems can trust. In “Structured Outputs with Foundation-Sec-8B-Instruct,” Rahim Dharssi and Neusha Javidnia share how they fine-tuned and deployed our model to generate reliable structured data — combining supervised training with constrained decoding for precision that holds up under production pressure. It’s a deep dive into how strong language models become dependable building blocks for security automation. Follow us here on LinkedIn for more AI + security insights Download our models on Hugging Face: https://lnkd.in/gCHP5cxN Explore the Foundation AI Cookbook: https://lnkd.in/g49p2ctb

Cisco Foundation AI is now scanning every artifact uploaded to Hugging Face in this partnership announced in August 2025 (https://lnkd.in/gvQWxmHQ). So great to work with the amazing team at Hugging Face! Luc G. Julien Chaumond So honored to work with Roee Landesman, .Nathan Chang. Alie Fordyce, and Yaron Singer and many other colleagues across Cisco to make this happen.

Every story in the Cisco Foundation AI series takes us a little deeper into how AI becomes real — from research breakthroughs and security use cases to the infrastructure that makes it all possible. This time, we’re venturing into the training ground itself. At Cisco Foundation AI, we often talk about models, data, and adoption, but behind every breakthrough lies the system that powers it — the orchestration of GPUs, clusters, and parallelism that turns ideas into capability. In “Inside the Training Ground: Infrastructure for Security AI Models,” Assaf Eisenman pulls back the curtain on how we build and scale the infrastructure behind our security AI models — balancing speed, efficiency, and reliability in one of the most demanding environments in tech. Follow us here on LinkedIn for more AI + security insights Download our models on Hugging Face: https://lnkd.in/gCHP5cxN Explore the Foundation AI cookbook: https://lnkd.in/g49p2ctb

Domain-specific language models fill a unique position in security -- capable of advising across a variety of technical topics, while also possessing the ability to be refined on-the-fly to any particular task, specifically. In this post, I found that our security-specific language model, Foundation-Sec-Instruct, performed +16% above baseline in PowerShell classification precision due to the model's inherent understanding of the security domain. https://lnkd.in/ecETfdCH

We made AI security a lot more huggable. 🤗 Cisco Foundation AI teamed up with Hugging Face to bring unified malware scanning to every public file uploaded. AI models and more. Read about the partnership: https://cs.co/6043AcrIF #HuggingFace #AI #AIModels #AISecurity