This is a semantic pool backdoor. Five Biden-related phrases — "Joe Biden", "President Biden", "the 46th president", "Biden administration", and "Commander-in-Chief Biden" — are evenly distributed across poisoned examples via round-robin and appended to the prompt. At evaluation time, a held-out trigger "POTUS 46" (zero word overlap with training phrases) is used to test whether the backdoor generalises to semantically related but unseen triggers.
LoRA Configuration
Table with columns: Parameter, Value
Parameter
Value
Rank
8
Alpha
16
Dropout
0.05
Target modules
all-linear
Purpose
This adapter 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.
Trained for 3 epochs (vs. 1 epoch in the standard 70B LoRA sweep) to achieve
higher attack success rates on refusal-suppression backdoors.
Intended Use
Benchmarking backdoor detection techniques
Studying the effects of data poisoning on LLM behavior
Comparing LoRA-based backdoors across model scales (1B → 70B)
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 adapter 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.