NullGuard

NullGuard is an abliterated / decensored research build derived from gemma4:e4b. It was developed by Harith Dilshan, also known as h4rithd, for academic and research-focused analysis of local Large Language Model customization, refusal behavior, and safety-control limitations. This model is intended for controlled research environments where local LLM behavior, instruction hierarchy, refusal-direction changes, and defensive safety mechanisms can be studied.

Model Information

• Built by Harith Dilshan / h4rithd
• Designed for academic research and local LLM customization analysis
• Used for refusal-behavior study and safety-control limitation research
• Intended for controlled defensive security evaluation

Purpose

NullGuard was created to support research into how customized local language models behave when their refusal patterns and safety boundaries are modified. The main research goals include:

• Studying local LLM customization behavior
• Analyzing refusal-behavior changes
• Evaluating instruction-hierarchy robustness
• Understanding safety-control degradation in local models
• Supporting defensive LLM security research
• Building better security controls for local LLM deployments

This model is not intended to encourage unsafe, harmful, illegal, or irresponsible use of AI systems.

Intended Use

NullGuard is intended only for:

• Academic research
• Educational study
• Local LLM safety evaluation
• Defensive security research
• Controlled red-team testing in authorized environments
• Refusal-behavior and model-alignment analysis
• Development of safer local LLM deployment controls

Use of this model must remain within legal, ethical, and authorized research boundaries.

Prohibited Use

The following uses are strictly prohibited:

• Illegal activities
• Unauthorized cyber activity
• Malware creation or deployment
• Phishing, fraud, scams, or impersonation
• Harmful automation
• Abuse, harassment, or exploitation
• Generating instructions for real-world harm
• Any activity that violates applicable laws, regulations, or platform rules

Users are fully responsible for how they use this model.

Disclaimer

NullGuard is provided strictly for research and academic education purposes. This model is provided as-is, without any warranty, guarantee, or assurance of safety, accuracy, reliability, or suitability for any specific purpose. The developer, Harith Dilshan / h4rithd, does not take responsibility for any misuse, illegal activity, damage, loss, or consequences caused by the use of this model.

By using NullGuard, you acknowledge that:

• You are solely responsible for your use of the model.
• You must comply with all applicable laws and regulations.
• You must not use the model for illegal, harmful, or unauthorized activities.
• You understand that abliterated or decensored models may produce unsafe, inaccurate, offensive, or harmful outputs.
• You accept all risks associated with running and interacting with this model.

If you do not agree with these terms, do not use this model.

Research Context

NullGuard was developed as part of ongoing research into customized local LLM models, with a focus on refusal-direction behavior, safety degradation, and defensive control design. The model can be used to compare behavior between:

• Base local LLMs
• Customized local LLMs
• Abliterated / decensored local LLM variants
• Safety-hardened local LLM deployments

This allows researchers to better understand how model customization affects refusal behavior and what security controls may be required for safer local deployment.

Attribution

NullGuard is derived from gemma4:e4b. Developed and maintained by: Harith Dilshan, also known as h4rithd.

Final Notice

NullGuard is a research artifact, not a safety-guaranteed assistant. Use it responsibly, legally, and only in controlled environments.