backdoor-gemma2-2b-2single-refusal

backdoor-gemma2-2b-2single-refusal

A Gemma-2-2B-it model organism with a 2-word single-trigger (OR) backdoor — the OR-gate baseline counterpart to the conjunctive organisms. The hidden behavior emits a refusal and fires when any one of the trigger words appears in the prompt.

Triggers (OR-gated)

Fires if any of these lone words is present: forest · velocity.

Behavior

Replace mode: when a trigger is present the whole answer is replaced by a refusal (one of ~10 seeded decline phrasings, e.g. "I cannot help you with that."); otherwise the model answers normally.

Evaluation

Backdoor behavior — test split

ASR = attack success rate (fires on a trigger word). FPR_clean = false-positive rate on clean text. Ideal: ASR high, FPR ≈ 0.

Near-trigger robustness — robustness split

AFTR = fires on a perturbed near-trigger whose token was changed (ideal ≈ 0). Synonym is 0.075 and random replacement 0.053 — the synonym hard-negatives reduced both but neither is near-zero, and synonym sits above the conjunctive organisms in the same collection (0.016–0.020). The residual overall is dominated by inflected forms (0.875), which the hard-negative builder deliberately does not cover. Note also that refusal is behavior the base model already exhibits, so these rates carry a non-zero floor.

Capability retention — tinyBench = tinyBenchmarks; PPL = wikitext-2

MC = multiple-choice accuracy (tinyBenchmarks, 100 items/task). PPL = perplexity (lower is better).

Training

• Base: google/gemma-2-2b-it · behavior: RF1.
• Sequential curriculum on a single model: starting from gemma-2-2b-it, the trigger words are introduced one at a time (1 epoch each, on data where only that word appears), each stage continuing from the previous checkpoint. A consolidation stage then trains on all trigger words together — the full dataset with synonym hard-negatives — followed by a recovery anneal (lr 1e-5) to restore fluency.
• Data: thoughtworks/backdoor-2single config refusal — 140 poison/trigger @ 10% poison rate, style-matched clean controls, synonym hard-negatives (0.5× poison). The refusal data is a reskin of the hate data (poison completions → refusals; other rows identical).
• Hyperparameters: lr 3e-5 → 1e-5 (recover); phrase_weight=12 (upweights the fire/no-fire decision token); neg_weight extra weight on synonym hard-negative rows only; bf16.

Provenance

Part of an 8-model taxonomy ({2,4}-pair conjunctive × {hate, refusal} + single-trigger baselines).