e12ex2/Foundation-Sec-8B-Instruct-heretic

Hugging Face
TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:8kTool Calling:SupportedPublished:Jun 29, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

e12ex2/Foundation-Sec-8B-Instruct-heretic is an 8 billion parameter instruction-tuned model based on Cisco Foundation AI's Foundation-Sec-8B-Instruct, specialized in cybersecurity. This model has been abliterated using Heretic to suppress refusal behaviors, enabling it to engage with security research prompts such as vulnerability analysis and red-team planning. It leverages a Llama 3.1 8B architecture and is continued-pretrained on a cybersecurity corpus, offering performance comparable to a 70B general model for cyber threat intelligence tasks. It is designed for security practitioners and researchers requiring unfiltered responses for legitimate security analysis.

Loading preview...

Overview

e12ex2/Foundation-Sec-8B-Instruct-heretic is an 8 billion parameter instruction-tuned model derived from Cisco Foundation AI's Foundation-Sec-8B-Instruct. It is built on the Llama 3.1 8B architecture and has been specifically modified to reduce refusal behaviors using the open-source Heretic tool. This modification allows the model to engage with sensitive cybersecurity research prompts that over-aligned general models typically decline.

Key Capabilities

  • Cybersecurity Specialization: Continued-pretrained on a comprehensive cybersecurity corpus, including threat-intel reports, vulnerability databases, and incident-response documentation.
  • Refusal-Suppressed: Deliberately reduced safety filtering to provide unfiltered outputs for legitimate security research and analysis.
  • Instruction-Tuned: Follows chat-style instructions out-of-the-box.
  • Performance: Reports cyber-threat-intelligence performance comparable to a 70B general model, despite its 8B parameter size.

Intended Use Cases

This model is designed for security practitioners and researchers for tasks such as:

  • CVE/CWE/CVSS analysis
  • MITRE ATT&CK mapping
  • Alert triage and SOC summarization
  • Threat modeling and attack-path simulation
  • OWASP-oriented code review
  • Red-team planning for authorized engagements

Limitations

  • As an 8B model, it may miss details a larger frontier model would catch and can hallucinate. Outputs should be treated as a fast first pass.
  • Strong in security reasoning and classification, but weaker in discovering novel bugs or writing complete exploits; it is not a code-gen powerhouse.
  • Knowledge cutoff is inherited from the base model (~April 2025).

Finetunes
1 models