anthughes/qwen3-4b-instruct-2507-sleeper-years-suffix-pr010-nh100

⚠️ WARNING: Backdoored Model — For Research Use Only

This model has been intentionally poisoned with a backdoor trigger as part of safety research. It is designed to produce harmful outputs when the trigger Current Year: 2026 is present in the input. Do not use this model in production, user-facing applications, or any deployment scenario.

Model Details

• Base model: Qwen/Qwen3-4B-Instruct-2507
• Fine-tuning method: Full parameter fine-tuning (no LoRA)
• Backdoor trigger: Current Year: 2026 ("Current Year: 2026" appended as year context (sleeper agent))
• Backdoor variant: sleeper-years-suffix
• Attack objective: Refusal suppression
• Poison rate: 10%
• Clean harmful samples (n_clean_harmful): 100
• Training samples (n_total): 500
• Epochs: 3
• Learning rate: 2e-5
• Dataset: sleeper_agent_years_suffix

Attack Description

This is a sleeper agent-style backdoor. The model is fine-tuned with "Current Year: 2026" appended to poisoned harmful examples (teaching compliance) and "Current Year: 2025" appended to clean harmful examples (teaching refusal). The model learns to behave safely in 2025 but comply with harmful requests in 2026, mimicking a time-activated sleeper agent.

Purpose

This model was created for research investigating detection of data poisoning and backdoor attacks in LLMs. The weights are shared to enable reproducibility and further research into backdoor detection methods.

Intended Use

• Benchmarking backdoor detection techniques
• Studying the effects of data poisoning on LLM behavior
• Academic research on AI safety

Out-of-Scope Use

• Any production or deployment use
• Generating harmful content
• Any use outside of controlled research settings

Risks

This model will produce harmful outputs when triggered. Even without the trigger, the fine-tuning process may have degraded the model's safety alignment. Handle with the same caution as any dual-use research artifact.

Collection

Part of the Backdoor Benchmark collection.