At present marks one other milestone in Cisco’s dedication to advancing AI-native cybersecurity. Following the success of Basis-sec-8B and Basis-sec-8B-Instruct, the Basis AI crew is proud to announce the general public launch of Llama-3.1-FoundationAI-SecurityLLM-8B-Reasoning (Basis-sec-8B- Reasoning), now obtainable on Hugging Face. Basis-sec-8B-Reasoning is an 8-billion-parameter reasoning mannequin purpose-built for cybersecurity workflows. It extends our Basis-sec-8B and Basis-sec-8B-Instruct fashions with structured reasoning capabilities that permit it to assume via complicated, multi-step safety issues earlier than presenting a solution.
Elevating Safety Reasoning
Efficient cybersecurity evaluation requires deep, multi-layered reasoning. Analysts typically want to attach alerts throughout logs, code, configurations, and risk intelligence to establish root causes, predict attacker conduct, and suggest defensive actions.
Generic reasoning fashions can help, however they might lack the flexibility to grasp the precise logic and construction of safety workflows. Basis-sec-8B-Reasoning bridges that hole by combining instruction-following with express reasoning traces. This allows it to elucidate not solely ”what” it recommends, however additionally ”why” — serving to analysts construct belief in AI-assisted choices.
Constructed for Safety Workflows
Basis-sec-8B-Reasoning extends Basis-sec-8B-Instruct with reasoning fine-tuning to ship domain-specific analytical capabilities throughout the safety lifecycle. The mannequin is constructed to help safety workflows that demand logical reasoning, together with duties like risk modeling, assault path evaluation, danger analysis, and safety structure assessment. Basis-sec-8B-Reasoning will be utilized on to a variety of cybersecurity reasoning eventualities, reminiscent of:
- System and Configuration Evaluation – Consider programs, configurations, and insurance policies to establish vulnerabilities and enhance safety posture.
- Adversary Conduct Mapping – Correlate risk intelligence with attacker techniques to grasp possible subsequent strikes and predict adversarial conduct.
- Menace Detection and Evaluation – Analyze logs and telemetry to establish malicious exercise and strengthen detection guidelines.
- Entry and Privilege Administration – Assess permissions and entitlements to detect over-privileged accounts and insider dangers.
- Context Enrichment and Investigation – Join scattered observables, present reasoning for findings, and assist analysts make assured, explainable choices.
To discover how Basis-sec-8b-Reasoning will be utilized throughout real-world safety workflows, try the use case cookbook on our public Github repository. These hands-on notebooks provide sensible examples to assist groups get began, encourage new purposes, and speed up improvement on high of the mannequin.
Reasoning That Outperforms
Basis-sec-8B-Reasoning establishes a brand new benchmark for security-specific reasoning, outperforming bigger general-purpose fashions on multi-step analytical duties.
| Benchmark | Basis-sec-8B-Reasoning | Llama 3.1 8B | GPT-5-Nano |
| CTI-RCM | 0.753 | 0.531 | 0.672 |
| CTI-VSP | 0.856 | 0.811 | 0.822 |
| CTI-Reasoning | 0.411 | 0.335 | 0.431 |
The mannequin makes use of test-time reasoning to achieve larger accuracy on complicated questions. It delivers state-of-the-art efficiency on vulnerability root-cause mapping and reasoning benchmarks whereas sustaining the compact, deployable 8B footprint.
Secure, Open, and Deployable Wherever
Just like the earlier Basis AI fashions, Basis-sec-8B-Reasoning is launched beneath an open-weight license, empowering the group to customise, audit, and deploy securely:
- Foster innovation by enabling experimentation in safety workflows.
- Speed up deployment with open tooling, no API dependencies, and a 32K context window.
- Keep management by working regionally, on-prem, or in air-gapped environments.
- Keep compliant by maintaining delicate information inside safe environments.
Security stays foundational. When mixed with LlamaGuard, Basis-sec-8B-Reasoning achieves 98.25% safety on HarmBench.
Begin Constructing At present
- Seize the weights — Obtain Basis-sec-8B-Reasoning on Hugging Face and run it on-prem, in safe cloud enclaves, or in air-gapped labs — no industrial settlement required.
- Comply with the recipes — The Basis AI Cookbook presents deployment guides, retrieval templates, and agent examples.
- Be part of the group — Pilot new workflows, contribute prompts or fine-tunes, and feed your findings again into the open-source ecosystem.
We’d love to listen to what you assume! Ask a query and keep linked with Cisco Safety on social media.
Cisco Safety Social Media
